xref: /dpdk/drivers/net/cxgbe/cxgbe_ethdev.c (revision a0163693bcc4f78f50c3471dbcacc2ca1394f71f)
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 <rte_ethdev_driver.h>
32 #include <rte_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 "t4_pci_id_tbl.h"
61 
62 #define CXGBE_TX_OFFLOADS (DEV_TX_OFFLOAD_VLAN_INSERT |\
63 			   DEV_TX_OFFLOAD_IPV4_CKSUM |\
64 			   DEV_TX_OFFLOAD_UDP_CKSUM |\
65 			   DEV_TX_OFFLOAD_TCP_CKSUM |\
66 			   DEV_TX_OFFLOAD_TCP_TSO)
67 
68 #define CXGBE_RX_OFFLOADS (DEV_RX_OFFLOAD_VLAN_STRIP |\
69 			   DEV_RX_OFFLOAD_CRC_STRIP |\
70 			   DEV_RX_OFFLOAD_IPV4_CKSUM |\
71 			   DEV_RX_OFFLOAD_JUMBO_FRAME |\
72 			   DEV_RX_OFFLOAD_UDP_CKSUM |\
73 			   DEV_RX_OFFLOAD_TCP_CKSUM)
74 
75 uint16_t cxgbe_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
76 			 uint16_t nb_pkts)
77 {
78 	struct sge_eth_txq *txq = (struct sge_eth_txq *)tx_queue;
79 	uint16_t pkts_sent, pkts_remain;
80 	uint16_t total_sent = 0;
81 	int ret = 0;
82 
83 	CXGBE_DEBUG_TX(adapter, "%s: txq = %p; tx_pkts = %p; nb_pkts = %d\n",
84 		       __func__, txq, tx_pkts, nb_pkts);
85 
86 	t4_os_lock(&txq->txq_lock);
87 	/* free up desc from already completed tx */
88 	reclaim_completed_tx(&txq->q);
89 	while (total_sent < nb_pkts) {
90 		pkts_remain = nb_pkts - total_sent;
91 
92 		for (pkts_sent = 0; pkts_sent < pkts_remain; pkts_sent++) {
93 			ret = t4_eth_xmit(txq, tx_pkts[total_sent + pkts_sent],
94 					  nb_pkts);
95 			if (ret < 0)
96 				break;
97 		}
98 		if (!pkts_sent)
99 			break;
100 		total_sent += pkts_sent;
101 		/* reclaim as much as possible */
102 		reclaim_completed_tx(&txq->q);
103 	}
104 
105 	t4_os_unlock(&txq->txq_lock);
106 	return total_sent;
107 }
108 
109 uint16_t cxgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
110 			 uint16_t nb_pkts)
111 {
112 	struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)rx_queue;
113 	unsigned int work_done;
114 
115 	CXGBE_DEBUG_RX(adapter, "%s: rxq->rspq.cntxt_id = %u; nb_pkts = %d\n",
116 		       __func__, rxq->rspq.cntxt_id, nb_pkts);
117 
118 	if (cxgbe_poll(&rxq->rspq, rx_pkts, (unsigned int)nb_pkts, &work_done))
119 		dev_err(adapter, "error in cxgbe poll\n");
120 
121 	CXGBE_DEBUG_RX(adapter, "%s: work_done = %u\n", __func__, work_done);
122 	return work_done;
123 }
124 
125 void cxgbe_dev_info_get(struct rte_eth_dev *eth_dev,
126 			struct rte_eth_dev_info *device_info)
127 {
128 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
129 	struct adapter *adapter = pi->adapter;
130 	int max_queues = adapter->sge.max_ethqsets / adapter->params.nports;
131 
132 	static const struct rte_eth_desc_lim cxgbe_desc_lim = {
133 		.nb_max = CXGBE_MAX_RING_DESC_SIZE,
134 		.nb_min = CXGBE_MIN_RING_DESC_SIZE,
135 		.nb_align = 1,
136 	};
137 
138 	device_info->min_rx_bufsize = CXGBE_MIN_RX_BUFSIZE;
139 	device_info->max_rx_pktlen = CXGBE_MAX_RX_PKTLEN;
140 	device_info->max_rx_queues = max_queues;
141 	device_info->max_tx_queues = max_queues;
142 	device_info->max_mac_addrs = 1;
143 	/* XXX: For now we support one MAC/port */
144 	device_info->max_vfs = adapter->params.arch.vfcount;
145 	device_info->max_vmdq_pools = 0; /* XXX: For now no support for VMDQ */
146 
147 	device_info->rx_queue_offload_capa = 0UL;
148 	device_info->rx_offload_capa = CXGBE_RX_OFFLOADS;
149 
150 	device_info->tx_queue_offload_capa = 0UL;
151 	device_info->tx_offload_capa = CXGBE_TX_OFFLOADS;
152 
153 	device_info->reta_size = pi->rss_size;
154 	device_info->hash_key_size = CXGBE_DEFAULT_RSS_KEY_LEN;
155 	device_info->flow_type_rss_offloads = CXGBE_RSS_HF_ALL;
156 
157 	device_info->rx_desc_lim = cxgbe_desc_lim;
158 	device_info->tx_desc_lim = cxgbe_desc_lim;
159 	cxgbe_get_speed_caps(pi, &device_info->speed_capa);
160 }
161 
162 void cxgbe_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
163 {
164 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
165 	struct adapter *adapter = pi->adapter;
166 
167 	t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
168 		      1, -1, 1, -1, false);
169 }
170 
171 void cxgbe_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
172 {
173 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
174 	struct adapter *adapter = pi->adapter;
175 
176 	t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
177 		      0, -1, 1, -1, false);
178 }
179 
180 void cxgbe_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
181 {
182 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
183 	struct adapter *adapter = pi->adapter;
184 
185 	/* TODO: address filters ?? */
186 
187 	t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
188 		      -1, 1, 1, -1, false);
189 }
190 
191 void cxgbe_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
192 {
193 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
194 	struct adapter *adapter = pi->adapter;
195 
196 	/* TODO: address filters ?? */
197 
198 	t4_set_rxmode(adapter, adapter->mbox, pi->viid, -1,
199 		      -1, 0, 1, -1, false);
200 }
201 
202 int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
203 			  int wait_to_complete)
204 {
205 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
206 	struct adapter *adapter = pi->adapter;
207 	struct sge *s = &adapter->sge;
208 	struct rte_eth_link new_link = { 0 };
209 	unsigned int i, work_done, budget = 32;
210 	u8 old_link = pi->link_cfg.link_ok;
211 
212 	for (i = 0; i < CXGBE_LINK_STATUS_POLL_CNT; i++) {
213 		cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
214 
215 		/* Exit if link status changed or always forced up */
216 		if (pi->link_cfg.link_ok != old_link || force_linkup(adapter))
217 			break;
218 
219 		if (!wait_to_complete)
220 			break;
221 
222 		rte_delay_ms(CXGBE_LINK_STATUS_POLL_MS);
223 	}
224 
225 	new_link.link_status = force_linkup(adapter) ?
226 			       ETH_LINK_UP : pi->link_cfg.link_ok;
227 	new_link.link_autoneg = pi->link_cfg.autoneg;
228 	new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
229 	new_link.link_speed = pi->link_cfg.speed;
230 
231 	return rte_eth_linkstatus_set(eth_dev, &new_link);
232 }
233 
234 /**
235  * Set device link up.
236  */
237 int cxgbe_dev_set_link_up(struct rte_eth_dev *dev)
238 {
239 	struct port_info *pi = (struct port_info *)(dev->data->dev_private);
240 	struct adapter *adapter = pi->adapter;
241 	unsigned int work_done, budget = 32;
242 	struct sge *s = &adapter->sge;
243 	int ret;
244 
245 	/* Flush all link events */
246 	cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
247 
248 	/* If link already up, nothing to do */
249 	if (pi->link_cfg.link_ok)
250 		return 0;
251 
252 	ret = cxgbe_set_link_status(pi, true);
253 	if (ret)
254 		return ret;
255 
256 	cxgbe_dev_link_update(dev, 1);
257 	return 0;
258 }
259 
260 /**
261  * Set device link down.
262  */
263 int cxgbe_dev_set_link_down(struct rte_eth_dev *dev)
264 {
265 	struct port_info *pi = (struct port_info *)(dev->data->dev_private);
266 	struct adapter *adapter = pi->adapter;
267 	unsigned int work_done, budget = 32;
268 	struct sge *s = &adapter->sge;
269 	int ret;
270 
271 	/* Flush all link events */
272 	cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
273 
274 	/* If link already down, nothing to do */
275 	if (!pi->link_cfg.link_ok)
276 		return 0;
277 
278 	ret = cxgbe_set_link_status(pi, false);
279 	if (ret)
280 		return ret;
281 
282 	cxgbe_dev_link_update(dev, 0);
283 	return 0;
284 }
285 
286 int cxgbe_dev_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
287 {
288 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
289 	struct adapter *adapter = pi->adapter;
290 	struct rte_eth_dev_info dev_info;
291 	int err;
292 	uint16_t new_mtu = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
293 
294 	cxgbe_dev_info_get(eth_dev, &dev_info);
295 
296 	/* Must accommodate at least ETHER_MIN_MTU */
297 	if ((new_mtu < ETHER_MIN_MTU) || (new_mtu > dev_info.max_rx_pktlen))
298 		return -EINVAL;
299 
300 	/* set to jumbo mode if needed */
301 	if (new_mtu > ETHER_MAX_LEN)
302 		eth_dev->data->dev_conf.rxmode.offloads |=
303 			DEV_RX_OFFLOAD_JUMBO_FRAME;
304 	else
305 		eth_dev->data->dev_conf.rxmode.offloads &=
306 			~DEV_RX_OFFLOAD_JUMBO_FRAME;
307 
308 	err = t4_set_rxmode(adapter, adapter->mbox, pi->viid, new_mtu, -1, -1,
309 			    -1, -1, true);
310 	if (!err)
311 		eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_mtu;
312 
313 	return err;
314 }
315 
316 /*
317  * Stop device.
318  */
319 void cxgbe_dev_close(struct rte_eth_dev *eth_dev)
320 {
321 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
322 	struct adapter *adapter = pi->adapter;
323 
324 	CXGBE_FUNC_TRACE();
325 
326 	if (!(adapter->flags & FULL_INIT_DONE))
327 		return;
328 
329 	cxgbe_down(pi);
330 
331 	/*
332 	 *  We clear queues only if both tx and rx path of the port
333 	 *  have been disabled
334 	 */
335 	t4_sge_eth_clear_queues(pi);
336 }
337 
338 /* Start the device.
339  * It returns 0 on success.
340  */
341 int cxgbe_dev_start(struct rte_eth_dev *eth_dev)
342 {
343 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
344 	struct adapter *adapter = pi->adapter;
345 	int err = 0, i;
346 
347 	CXGBE_FUNC_TRACE();
348 
349 	/*
350 	 * If we don't have a connection to the firmware there's nothing we
351 	 * can do.
352 	 */
353 	if (!(adapter->flags & FW_OK)) {
354 		err = -ENXIO;
355 		goto out;
356 	}
357 
358 	if (!(adapter->flags & FULL_INIT_DONE)) {
359 		err = cxgbe_up(adapter);
360 		if (err < 0)
361 			goto out;
362 	}
363 
364 	cxgbe_enable_rx_queues(pi);
365 
366 	err = setup_rss(pi);
367 	if (err)
368 		goto out;
369 
370 	for (i = 0; i < pi->n_tx_qsets; i++) {
371 		err = cxgbe_dev_tx_queue_start(eth_dev, i);
372 		if (err)
373 			goto out;
374 	}
375 
376 	for (i = 0; i < pi->n_rx_qsets; i++) {
377 		err = cxgbe_dev_rx_queue_start(eth_dev, i);
378 		if (err)
379 			goto out;
380 	}
381 
382 	err = link_start(pi);
383 	if (err)
384 		goto out;
385 
386 out:
387 	return err;
388 }
389 
390 /*
391  * Stop device: disable rx and tx functions to allow for reconfiguring.
392  */
393 void cxgbe_dev_stop(struct rte_eth_dev *eth_dev)
394 {
395 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
396 	struct adapter *adapter = pi->adapter;
397 
398 	CXGBE_FUNC_TRACE();
399 
400 	if (!(adapter->flags & FULL_INIT_DONE))
401 		return;
402 
403 	cxgbe_down(pi);
404 
405 	/*
406 	 *  We clear queues only if both tx and rx path of the port
407 	 *  have been disabled
408 	 */
409 	t4_sge_eth_clear_queues(pi);
410 }
411 
412 int cxgbe_dev_configure(struct rte_eth_dev *eth_dev)
413 {
414 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
415 	struct adapter *adapter = pi->adapter;
416 	uint64_t configured_offloads;
417 	int err;
418 
419 	CXGBE_FUNC_TRACE();
420 	configured_offloads = eth_dev->data->dev_conf.rxmode.offloads;
421 
422 	/* KEEP_CRC offload flag is not supported by PMD
423 	 * can remove the below block when DEV_RX_OFFLOAD_CRC_STRIP removed
424 	 */
425 	if (rte_eth_dev_must_keep_crc(configured_offloads)) {
426 		dev_info(adapter, "can't disable hw crc strip\n");
427 		eth_dev->data->dev_conf.rxmode.offloads |=
428 			DEV_RX_OFFLOAD_CRC_STRIP;
429 	}
430 
431 	if (!(adapter->flags & FW_QUEUE_BOUND)) {
432 		err = setup_sge_fwevtq(adapter);
433 		if (err)
434 			return err;
435 		adapter->flags |= FW_QUEUE_BOUND;
436 		if (is_pf4(adapter)) {
437 			err = setup_sge_ctrl_txq(adapter);
438 			if (err)
439 				return err;
440 		}
441 	}
442 
443 	err = cfg_queue_count(eth_dev);
444 	if (err)
445 		return err;
446 
447 	return 0;
448 }
449 
450 int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
451 {
452 	int ret;
453 	struct sge_eth_txq *txq = (struct sge_eth_txq *)
454 				  (eth_dev->data->tx_queues[tx_queue_id]);
455 
456 	dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
457 
458 	ret = t4_sge_eth_txq_start(txq);
459 	if (ret == 0)
460 		eth_dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
461 
462 	return ret;
463 }
464 
465 int cxgbe_dev_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
466 {
467 	int ret;
468 	struct sge_eth_txq *txq = (struct sge_eth_txq *)
469 				  (eth_dev->data->tx_queues[tx_queue_id]);
470 
471 	dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
472 
473 	ret = t4_sge_eth_txq_stop(txq);
474 	if (ret == 0)
475 		eth_dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
476 
477 	return ret;
478 }
479 
480 int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
481 			     uint16_t queue_idx, uint16_t nb_desc,
482 			     unsigned int socket_id,
483 			     const struct rte_eth_txconf *tx_conf __rte_unused)
484 {
485 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
486 	struct adapter *adapter = pi->adapter;
487 	struct sge *s = &adapter->sge;
488 	struct sge_eth_txq *txq = &s->ethtxq[pi->first_qset + queue_idx];
489 	int err = 0;
490 	unsigned int temp_nb_desc;
491 
492 	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",
493 		  __func__, eth_dev->data->nb_tx_queues, queue_idx, nb_desc,
494 		  socket_id, pi->first_qset);
495 
496 	/*  Free up the existing queue  */
497 	if (eth_dev->data->tx_queues[queue_idx]) {
498 		cxgbe_dev_tx_queue_release(eth_dev->data->tx_queues[queue_idx]);
499 		eth_dev->data->tx_queues[queue_idx] = NULL;
500 	}
501 
502 	eth_dev->data->tx_queues[queue_idx] = (void *)txq;
503 
504 	/* Sanity Checking
505 	 *
506 	 * nb_desc should be > 1023 and <= CXGBE_MAX_RING_DESC_SIZE
507 	 */
508 	temp_nb_desc = nb_desc;
509 	if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
510 		dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
511 			 __func__, CXGBE_MIN_RING_DESC_SIZE,
512 			 CXGBE_DEFAULT_TX_DESC_SIZE);
513 		temp_nb_desc = CXGBE_DEFAULT_TX_DESC_SIZE;
514 	} else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
515 		dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
516 			__func__, CXGBE_MIN_RING_DESC_SIZE,
517 			CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_TX_DESC_SIZE);
518 		return -(EINVAL);
519 	}
520 
521 	txq->q.size = temp_nb_desc;
522 
523 	err = t4_sge_alloc_eth_txq(adapter, txq, eth_dev, queue_idx,
524 				   s->fw_evtq.cntxt_id, socket_id);
525 
526 	dev_debug(adapter, "%s: txq->q.cntxt_id= %u txq->q.abs_id= %u err = %d\n",
527 		  __func__, txq->q.cntxt_id, txq->q.abs_id, err);
528 	return err;
529 }
530 
531 void cxgbe_dev_tx_queue_release(void *q)
532 {
533 	struct sge_eth_txq *txq = (struct sge_eth_txq *)q;
534 
535 	if (txq) {
536 		struct port_info *pi = (struct port_info *)
537 				       (txq->eth_dev->data->dev_private);
538 		struct adapter *adap = pi->adapter;
539 
540 		dev_debug(adapter, "%s: pi->port_id = %d; tx_queue_id = %d\n",
541 			  __func__, pi->port_id, txq->q.cntxt_id);
542 
543 		t4_sge_eth_txq_release(adap, txq);
544 	}
545 }
546 
547 int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
548 {
549 	int ret;
550 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
551 	struct adapter *adap = pi->adapter;
552 	struct sge_rspq *q;
553 
554 	dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
555 		  __func__, pi->port_id, rx_queue_id);
556 
557 	q = eth_dev->data->rx_queues[rx_queue_id];
558 
559 	ret = t4_sge_eth_rxq_start(adap, q);
560 	if (ret == 0)
561 		eth_dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
562 
563 	return ret;
564 }
565 
566 int cxgbe_dev_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
567 {
568 	int ret;
569 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
570 	struct adapter *adap = pi->adapter;
571 	struct sge_rspq *q;
572 
573 	dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
574 		  __func__, pi->port_id, rx_queue_id);
575 
576 	q = eth_dev->data->rx_queues[rx_queue_id];
577 	ret = t4_sge_eth_rxq_stop(adap, q);
578 	if (ret == 0)
579 		eth_dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
580 
581 	return ret;
582 }
583 
584 int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
585 			     uint16_t queue_idx, uint16_t nb_desc,
586 			     unsigned int socket_id,
587 			     const struct rte_eth_rxconf *rx_conf __rte_unused,
588 			     struct rte_mempool *mp)
589 {
590 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
591 	struct adapter *adapter = pi->adapter;
592 	struct sge *s = &adapter->sge;
593 	struct sge_eth_rxq *rxq = &s->ethrxq[pi->first_qset + queue_idx];
594 	int err = 0;
595 	int msi_idx = 0;
596 	unsigned int temp_nb_desc;
597 	struct rte_eth_dev_info dev_info;
598 	unsigned int pkt_len = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
599 
600 	dev_debug(adapter, "%s: eth_dev->data->nb_rx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; mp = %p\n",
601 		  __func__, eth_dev->data->nb_rx_queues, queue_idx, nb_desc,
602 		  socket_id, mp);
603 
604 	cxgbe_dev_info_get(eth_dev, &dev_info);
605 
606 	/* Must accommodate at least ETHER_MIN_MTU */
607 	if ((pkt_len < dev_info.min_rx_bufsize) ||
608 	    (pkt_len > dev_info.max_rx_pktlen)) {
609 		dev_err(adap, "%s: max pkt len must be > %d and <= %d\n",
610 			__func__, dev_info.min_rx_bufsize,
611 			dev_info.max_rx_pktlen);
612 		return -EINVAL;
613 	}
614 
615 	/*  Free up the existing queue  */
616 	if (eth_dev->data->rx_queues[queue_idx]) {
617 		cxgbe_dev_rx_queue_release(eth_dev->data->rx_queues[queue_idx]);
618 		eth_dev->data->rx_queues[queue_idx] = NULL;
619 	}
620 
621 	eth_dev->data->rx_queues[queue_idx] = (void *)rxq;
622 
623 	/* Sanity Checking
624 	 *
625 	 * nb_desc should be > 0 and <= CXGBE_MAX_RING_DESC_SIZE
626 	 */
627 	temp_nb_desc = nb_desc;
628 	if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
629 		dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
630 			 __func__, CXGBE_MIN_RING_DESC_SIZE,
631 			 CXGBE_DEFAULT_RX_DESC_SIZE);
632 		temp_nb_desc = CXGBE_DEFAULT_RX_DESC_SIZE;
633 	} else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
634 		dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
635 			__func__, CXGBE_MIN_RING_DESC_SIZE,
636 			CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_RX_DESC_SIZE);
637 		return -(EINVAL);
638 	}
639 
640 	rxq->rspq.size = temp_nb_desc;
641 	if ((&rxq->fl) != NULL)
642 		rxq->fl.size = temp_nb_desc;
643 
644 	/* Set to jumbo mode if necessary */
645 	if (pkt_len > ETHER_MAX_LEN)
646 		eth_dev->data->dev_conf.rxmode.offloads |=
647 			DEV_RX_OFFLOAD_JUMBO_FRAME;
648 	else
649 		eth_dev->data->dev_conf.rxmode.offloads &=
650 			~DEV_RX_OFFLOAD_JUMBO_FRAME;
651 
652 	err = t4_sge_alloc_rxq(adapter, &rxq->rspq, false, eth_dev, msi_idx,
653 			       &rxq->fl, t4_ethrx_handler,
654 			       is_pf4(adapter) ?
655 			       t4_get_tp_ch_map(adapter, pi->tx_chan) : 0, mp,
656 			       queue_idx, socket_id);
657 
658 	dev_debug(adapter, "%s: err = %d; port_id = %d; cntxt_id = %u; abs_id = %u\n",
659 		  __func__, err, pi->port_id, rxq->rspq.cntxt_id,
660 		  rxq->rspq.abs_id);
661 	return err;
662 }
663 
664 void cxgbe_dev_rx_queue_release(void *q)
665 {
666 	struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)q;
667 	struct sge_rspq *rq = &rxq->rspq;
668 
669 	if (rq) {
670 		struct port_info *pi = (struct port_info *)
671 				       (rq->eth_dev->data->dev_private);
672 		struct adapter *adap = pi->adapter;
673 
674 		dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
675 			  __func__, pi->port_id, rxq->rspq.cntxt_id);
676 
677 		t4_sge_eth_rxq_release(adap, rxq);
678 	}
679 }
680 
681 /*
682  * Get port statistics.
683  */
684 static int cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
685 				struct rte_eth_stats *eth_stats)
686 {
687 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
688 	struct adapter *adapter = pi->adapter;
689 	struct sge *s = &adapter->sge;
690 	struct port_stats ps;
691 	unsigned int i;
692 
693 	cxgbe_stats_get(pi, &ps);
694 
695 	/* RX Stats */
696 	eth_stats->imissed  = ps.rx_ovflow0 + ps.rx_ovflow1 +
697 			      ps.rx_ovflow2 + ps.rx_ovflow3 +
698 			      ps.rx_trunc0 + ps.rx_trunc1 +
699 			      ps.rx_trunc2 + ps.rx_trunc3;
700 	eth_stats->ierrors  = ps.rx_symbol_err + ps.rx_fcs_err +
701 			      ps.rx_jabber + ps.rx_too_long + ps.rx_runt +
702 			      ps.rx_len_err;
703 
704 	/* TX Stats */
705 	eth_stats->opackets = ps.tx_frames;
706 	eth_stats->obytes   = ps.tx_octets;
707 	eth_stats->oerrors  = ps.tx_error_frames;
708 
709 	for (i = 0; i < pi->n_rx_qsets; i++) {
710 		struct sge_eth_rxq *rxq =
711 			&s->ethrxq[pi->first_qset + i];
712 
713 		eth_stats->q_ipackets[i] = rxq->stats.pkts;
714 		eth_stats->q_ibytes[i] = rxq->stats.rx_bytes;
715 		eth_stats->ipackets += eth_stats->q_ipackets[i];
716 		eth_stats->ibytes += eth_stats->q_ibytes[i];
717 	}
718 
719 	for (i = 0; i < pi->n_tx_qsets; i++) {
720 		struct sge_eth_txq *txq =
721 			&s->ethtxq[pi->first_qset + i];
722 
723 		eth_stats->q_opackets[i] = txq->stats.pkts;
724 		eth_stats->q_obytes[i] = txq->stats.tx_bytes;
725 		eth_stats->q_errors[i] = txq->stats.mapping_err;
726 	}
727 	return 0;
728 }
729 
730 /*
731  * Reset port statistics.
732  */
733 static void cxgbe_dev_stats_reset(struct rte_eth_dev *eth_dev)
734 {
735 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
736 	struct adapter *adapter = pi->adapter;
737 	struct sge *s = &adapter->sge;
738 	unsigned int i;
739 
740 	cxgbe_stats_reset(pi);
741 	for (i = 0; i < pi->n_rx_qsets; i++) {
742 		struct sge_eth_rxq *rxq =
743 			&s->ethrxq[pi->first_qset + i];
744 
745 		rxq->stats.pkts = 0;
746 		rxq->stats.rx_bytes = 0;
747 	}
748 	for (i = 0; i < pi->n_tx_qsets; i++) {
749 		struct sge_eth_txq *txq =
750 			&s->ethtxq[pi->first_qset + i];
751 
752 		txq->stats.pkts = 0;
753 		txq->stats.tx_bytes = 0;
754 		txq->stats.mapping_err = 0;
755 	}
756 }
757 
758 static int cxgbe_flow_ctrl_get(struct rte_eth_dev *eth_dev,
759 			       struct rte_eth_fc_conf *fc_conf)
760 {
761 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
762 	struct link_config *lc = &pi->link_cfg;
763 	int rx_pause, tx_pause;
764 
765 	fc_conf->autoneg = lc->fc & PAUSE_AUTONEG;
766 	rx_pause = lc->fc & PAUSE_RX;
767 	tx_pause = lc->fc & PAUSE_TX;
768 
769 	if (rx_pause && tx_pause)
770 		fc_conf->mode = RTE_FC_FULL;
771 	else if (rx_pause)
772 		fc_conf->mode = RTE_FC_RX_PAUSE;
773 	else if (tx_pause)
774 		fc_conf->mode = RTE_FC_TX_PAUSE;
775 	else
776 		fc_conf->mode = RTE_FC_NONE;
777 	return 0;
778 }
779 
780 static int cxgbe_flow_ctrl_set(struct rte_eth_dev *eth_dev,
781 			       struct rte_eth_fc_conf *fc_conf)
782 {
783 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
784 	struct adapter *adapter = pi->adapter;
785 	struct link_config *lc = &pi->link_cfg;
786 
787 	if (lc->pcaps & FW_PORT_CAP32_ANEG) {
788 		if (fc_conf->autoneg)
789 			lc->requested_fc |= PAUSE_AUTONEG;
790 		else
791 			lc->requested_fc &= ~PAUSE_AUTONEG;
792 	}
793 
794 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
795 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
796 		lc->requested_fc |= PAUSE_RX;
797 	else
798 		lc->requested_fc &= ~PAUSE_RX;
799 
800 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
801 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
802 		lc->requested_fc |= PAUSE_TX;
803 	else
804 		lc->requested_fc &= ~PAUSE_TX;
805 
806 	return t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
807 			     &pi->link_cfg);
808 }
809 
810 const uint32_t *
811 cxgbe_dev_supported_ptypes_get(struct rte_eth_dev *eth_dev)
812 {
813 	static const uint32_t ptypes[] = {
814 		RTE_PTYPE_L3_IPV4,
815 		RTE_PTYPE_L3_IPV6,
816 		RTE_PTYPE_UNKNOWN
817 	};
818 
819 	if (eth_dev->rx_pkt_burst == cxgbe_recv_pkts)
820 		return ptypes;
821 	return NULL;
822 }
823 
824 /* Update RSS hash configuration
825  */
826 static int cxgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
827 				     struct rte_eth_rss_conf *rss_conf)
828 {
829 	struct port_info *pi = (struct port_info *)(dev->data->dev_private);
830 	struct adapter *adapter = pi->adapter;
831 	int err;
832 
833 	err = cxgbe_write_rss_conf(pi, rss_conf->rss_hf);
834 	if (err)
835 		return err;
836 
837 	pi->rss_hf = rss_conf->rss_hf;
838 
839 	if (rss_conf->rss_key) {
840 		u32 key[10], mod_key[10];
841 		int i, j;
842 
843 		memcpy(key, rss_conf->rss_key, CXGBE_DEFAULT_RSS_KEY_LEN);
844 
845 		for (i = 9, j = 0; i >= 0; i--, j++)
846 			mod_key[j] = cpu_to_be32(key[i]);
847 
848 		t4_write_rss_key(adapter, mod_key, -1);
849 	}
850 
851 	return 0;
852 }
853 
854 /* Get RSS hash configuration
855  */
856 static int cxgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
857 				       struct rte_eth_rss_conf *rss_conf)
858 {
859 	struct port_info *pi = (struct port_info *)(dev->data->dev_private);
860 	struct adapter *adapter = pi->adapter;
861 	u64 rss_hf = 0;
862 	u64 flags = 0;
863 	int err;
864 
865 	err = t4_read_config_vi_rss(adapter, adapter->mbox, pi->viid,
866 				    &flags, NULL);
867 
868 	if (err)
869 		return err;
870 
871 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) {
872 		rss_hf |= CXGBE_RSS_HF_TCP_IPV6_MASK;
873 		if (flags & F_FW_RSS_VI_CONFIG_CMD_UDPEN)
874 			rss_hf |= CXGBE_RSS_HF_UDP_IPV6_MASK;
875 	}
876 
877 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
878 		rss_hf |= CXGBE_RSS_HF_IPV6_MASK;
879 
880 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) {
881 		rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
882 		if (flags & F_FW_RSS_VI_CONFIG_CMD_UDPEN)
883 			rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
884 	}
885 
886 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
887 		rss_hf |= CXGBE_RSS_HF_IPV4_MASK;
888 
889 	rss_conf->rss_hf = rss_hf;
890 
891 	if (rss_conf->rss_key) {
892 		u32 key[10], mod_key[10];
893 		int i, j;
894 
895 		t4_read_rss_key(adapter, key);
896 
897 		for (i = 9, j = 0; i >= 0; i--, j++)
898 			mod_key[j] = be32_to_cpu(key[i]);
899 
900 		memcpy(rss_conf->rss_key, mod_key, CXGBE_DEFAULT_RSS_KEY_LEN);
901 	}
902 
903 	return 0;
904 }
905 
906 static int cxgbe_get_eeprom_length(struct rte_eth_dev *dev)
907 {
908 	RTE_SET_USED(dev);
909 	return EEPROMSIZE;
910 }
911 
912 /**
913  * eeprom_ptov - translate a physical EEPROM address to virtual
914  * @phys_addr: the physical EEPROM address
915  * @fn: the PCI function number
916  * @sz: size of function-specific area
917  *
918  * Translate a physical EEPROM address to virtual.  The first 1K is
919  * accessed through virtual addresses starting at 31K, the rest is
920  * accessed through virtual addresses starting at 0.
921  *
922  * The mapping is as follows:
923  * [0..1K) -> [31K..32K)
924  * [1K..1K+A) -> [31K-A..31K)
925  * [1K+A..ES) -> [0..ES-A-1K)
926  *
927  * where A = @fn * @sz, and ES = EEPROM size.
928  */
929 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
930 {
931 	fn *= sz;
932 	if (phys_addr < 1024)
933 		return phys_addr + (31 << 10);
934 	if (phys_addr < 1024 + fn)
935 		return fn + phys_addr - 1024;
936 	if (phys_addr < EEPROMSIZE)
937 		return phys_addr - 1024 - fn;
938 	if (phys_addr < EEPROMVSIZE)
939 		return phys_addr - 1024;
940 	return -EINVAL;
941 }
942 
943 /* The next two routines implement eeprom read/write from physical addresses.
944  */
945 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
946 {
947 	int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
948 
949 	if (vaddr >= 0)
950 		vaddr = t4_seeprom_read(adap, vaddr, v);
951 	return vaddr < 0 ? vaddr : 0;
952 }
953 
954 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
955 {
956 	int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
957 
958 	if (vaddr >= 0)
959 		vaddr = t4_seeprom_write(adap, vaddr, v);
960 	return vaddr < 0 ? vaddr : 0;
961 }
962 
963 #define EEPROM_MAGIC 0x38E2F10C
964 
965 static int cxgbe_get_eeprom(struct rte_eth_dev *dev,
966 			    struct rte_dev_eeprom_info *e)
967 {
968 	struct port_info *pi = (struct port_info *)(dev->data->dev_private);
969 	struct adapter *adapter = pi->adapter;
970 	u32 i, err = 0;
971 	u8 *buf = rte_zmalloc(NULL, EEPROMSIZE, 0);
972 
973 	if (!buf)
974 		return -ENOMEM;
975 
976 	e->magic = EEPROM_MAGIC;
977 	for (i = e->offset & ~3; !err && i < e->offset + e->length; i += 4)
978 		err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
979 
980 	if (!err)
981 		rte_memcpy(e->data, buf + e->offset, e->length);
982 	rte_free(buf);
983 	return err;
984 }
985 
986 static int cxgbe_set_eeprom(struct rte_eth_dev *dev,
987 			    struct rte_dev_eeprom_info *eeprom)
988 {
989 	struct port_info *pi = (struct port_info *)(dev->data->dev_private);
990 	struct adapter *adapter = pi->adapter;
991 	u8 *buf;
992 	int err = 0;
993 	u32 aligned_offset, aligned_len, *p;
994 
995 	if (eeprom->magic != EEPROM_MAGIC)
996 		return -EINVAL;
997 
998 	aligned_offset = eeprom->offset & ~3;
999 	aligned_len = (eeprom->length + (eeprom->offset & 3) + 3) & ~3;
1000 
1001 	if (adapter->pf > 0) {
1002 		u32 start = 1024 + adapter->pf * EEPROMPFSIZE;
1003 
1004 		if (aligned_offset < start ||
1005 		    aligned_offset + aligned_len > start + EEPROMPFSIZE)
1006 			return -EPERM;
1007 	}
1008 
1009 	if (aligned_offset != eeprom->offset || aligned_len != eeprom->length) {
1010 		/* RMW possibly needed for first or last words.
1011 		 */
1012 		buf = rte_zmalloc(NULL, aligned_len, 0);
1013 		if (!buf)
1014 			return -ENOMEM;
1015 		err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
1016 		if (!err && aligned_len > 4)
1017 			err = eeprom_rd_phys(adapter,
1018 					     aligned_offset + aligned_len - 4,
1019 					     (u32 *)&buf[aligned_len - 4]);
1020 		if (err)
1021 			goto out;
1022 		rte_memcpy(buf + (eeprom->offset & 3), eeprom->data,
1023 			   eeprom->length);
1024 	} else {
1025 		buf = eeprom->data;
1026 	}
1027 
1028 	err = t4_seeprom_wp(adapter, false);
1029 	if (err)
1030 		goto out;
1031 
1032 	for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
1033 		err = eeprom_wr_phys(adapter, aligned_offset, *p);
1034 		aligned_offset += 4;
1035 	}
1036 
1037 	if (!err)
1038 		err = t4_seeprom_wp(adapter, true);
1039 out:
1040 	if (buf != eeprom->data)
1041 		rte_free(buf);
1042 	return err;
1043 }
1044 
1045 static int cxgbe_get_regs_len(struct rte_eth_dev *eth_dev)
1046 {
1047 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
1048 	struct adapter *adapter = pi->adapter;
1049 
1050 	return t4_get_regs_len(adapter) / sizeof(uint32_t);
1051 }
1052 
1053 static int cxgbe_get_regs(struct rte_eth_dev *eth_dev,
1054 			  struct rte_dev_reg_info *regs)
1055 {
1056 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
1057 	struct adapter *adapter = pi->adapter;
1058 
1059 	regs->version = CHELSIO_CHIP_VERSION(adapter->params.chip) |
1060 		(CHELSIO_CHIP_RELEASE(adapter->params.chip) << 10) |
1061 		(1 << 16);
1062 
1063 	if (regs->data == NULL) {
1064 		regs->length = cxgbe_get_regs_len(eth_dev);
1065 		regs->width = sizeof(uint32_t);
1066 
1067 		return 0;
1068 	}
1069 
1070 	t4_get_regs(adapter, regs->data, (regs->length * sizeof(uint32_t)));
1071 
1072 	return 0;
1073 }
1074 
1075 int cxgbe_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *addr)
1076 {
1077 	struct port_info *pi = (struct port_info *)(dev->data->dev_private);
1078 	struct adapter *adapter = pi->adapter;
1079 	int ret;
1080 
1081 	ret = t4_change_mac(adapter, adapter->mbox, pi->viid,
1082 			    pi->xact_addr_filt, (u8 *)addr, true, true);
1083 	if (ret < 0) {
1084 		dev_err(adapter, "failed to set mac addr; err = %d\n",
1085 			ret);
1086 		return ret;
1087 	}
1088 	pi->xact_addr_filt = ret;
1089 	return 0;
1090 }
1091 
1092 static const struct eth_dev_ops cxgbe_eth_dev_ops = {
1093 	.dev_start		= cxgbe_dev_start,
1094 	.dev_stop		= cxgbe_dev_stop,
1095 	.dev_close		= cxgbe_dev_close,
1096 	.promiscuous_enable	= cxgbe_dev_promiscuous_enable,
1097 	.promiscuous_disable	= cxgbe_dev_promiscuous_disable,
1098 	.allmulticast_enable	= cxgbe_dev_allmulticast_enable,
1099 	.allmulticast_disable	= cxgbe_dev_allmulticast_disable,
1100 	.dev_configure		= cxgbe_dev_configure,
1101 	.dev_infos_get		= cxgbe_dev_info_get,
1102 	.dev_supported_ptypes_get = cxgbe_dev_supported_ptypes_get,
1103 	.link_update		= cxgbe_dev_link_update,
1104 	.dev_set_link_up        = cxgbe_dev_set_link_up,
1105 	.dev_set_link_down      = cxgbe_dev_set_link_down,
1106 	.mtu_set		= cxgbe_dev_mtu_set,
1107 	.tx_queue_setup         = cxgbe_dev_tx_queue_setup,
1108 	.tx_queue_start		= cxgbe_dev_tx_queue_start,
1109 	.tx_queue_stop		= cxgbe_dev_tx_queue_stop,
1110 	.tx_queue_release	= cxgbe_dev_tx_queue_release,
1111 	.rx_queue_setup         = cxgbe_dev_rx_queue_setup,
1112 	.rx_queue_start		= cxgbe_dev_rx_queue_start,
1113 	.rx_queue_stop		= cxgbe_dev_rx_queue_stop,
1114 	.rx_queue_release	= cxgbe_dev_rx_queue_release,
1115 	.filter_ctrl            = cxgbe_dev_filter_ctrl,
1116 	.stats_get		= cxgbe_dev_stats_get,
1117 	.stats_reset		= cxgbe_dev_stats_reset,
1118 	.flow_ctrl_get		= cxgbe_flow_ctrl_get,
1119 	.flow_ctrl_set		= cxgbe_flow_ctrl_set,
1120 	.get_eeprom_length	= cxgbe_get_eeprom_length,
1121 	.get_eeprom		= cxgbe_get_eeprom,
1122 	.set_eeprom		= cxgbe_set_eeprom,
1123 	.get_reg		= cxgbe_get_regs,
1124 	.rss_hash_update	= cxgbe_dev_rss_hash_update,
1125 	.rss_hash_conf_get	= cxgbe_dev_rss_hash_conf_get,
1126 	.mac_addr_set		= cxgbe_mac_addr_set,
1127 };
1128 
1129 /*
1130  * Initialize driver
1131  * It returns 0 on success.
1132  */
1133 static int eth_cxgbe_dev_init(struct rte_eth_dev *eth_dev)
1134 {
1135 	struct rte_pci_device *pci_dev;
1136 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
1137 	struct adapter *adapter = NULL;
1138 	char name[RTE_ETH_NAME_MAX_LEN];
1139 	int err = 0;
1140 
1141 	CXGBE_FUNC_TRACE();
1142 
1143 	eth_dev->dev_ops = &cxgbe_eth_dev_ops;
1144 	eth_dev->rx_pkt_burst = &cxgbe_recv_pkts;
1145 	eth_dev->tx_pkt_burst = &cxgbe_xmit_pkts;
1146 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1147 
1148 	/* for secondary processes, we attach to ethdevs allocated by primary
1149 	 * and do minimal initialization.
1150 	 */
1151 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1152 		int i;
1153 
1154 		for (i = 1; i < MAX_NPORTS; i++) {
1155 			struct rte_eth_dev *rest_eth_dev;
1156 			char namei[RTE_ETH_NAME_MAX_LEN];
1157 
1158 			snprintf(namei, sizeof(namei), "%s_%d",
1159 				 pci_dev->device.name, i);
1160 			rest_eth_dev = rte_eth_dev_attach_secondary(namei);
1161 			if (rest_eth_dev) {
1162 				rest_eth_dev->device = &pci_dev->device;
1163 				rest_eth_dev->dev_ops =
1164 					eth_dev->dev_ops;
1165 				rest_eth_dev->rx_pkt_burst =
1166 					eth_dev->rx_pkt_burst;
1167 				rest_eth_dev->tx_pkt_burst =
1168 					eth_dev->tx_pkt_burst;
1169 				rte_eth_dev_probing_finish(rest_eth_dev);
1170 			}
1171 		}
1172 		return 0;
1173 	}
1174 
1175 	snprintf(name, sizeof(name), "cxgbeadapter%d", eth_dev->data->port_id);
1176 	adapter = rte_zmalloc(name, sizeof(*adapter), 0);
1177 	if (!adapter)
1178 		return -1;
1179 
1180 	adapter->use_unpacked_mode = 1;
1181 	adapter->regs = (void *)pci_dev->mem_resource[0].addr;
1182 	if (!adapter->regs) {
1183 		dev_err(adapter, "%s: cannot map device registers\n", __func__);
1184 		err = -ENOMEM;
1185 		goto out_free_adapter;
1186 	}
1187 	adapter->pdev = pci_dev;
1188 	adapter->eth_dev = eth_dev;
1189 	pi->adapter = adapter;
1190 
1191 	err = cxgbe_probe(adapter);
1192 	if (err) {
1193 		dev_err(adapter, "%s: cxgbe probe failed with err %d\n",
1194 			__func__, err);
1195 		goto out_free_adapter;
1196 	}
1197 
1198 	return 0;
1199 
1200 out_free_adapter:
1201 	rte_free(adapter);
1202 	return err;
1203 }
1204 
1205 static int eth_cxgbe_dev_uninit(struct rte_eth_dev *eth_dev)
1206 {
1207 	struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
1208 	struct adapter *adap = pi->adapter;
1209 
1210 	/* Free up other ports and all resources */
1211 	cxgbe_close(adap);
1212 	return 0;
1213 }
1214 
1215 static int eth_cxgbe_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1216 	struct rte_pci_device *pci_dev)
1217 {
1218 	return rte_eth_dev_pci_generic_probe(pci_dev,
1219 		sizeof(struct port_info), eth_cxgbe_dev_init);
1220 }
1221 
1222 static int eth_cxgbe_pci_remove(struct rte_pci_device *pci_dev)
1223 {
1224 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_cxgbe_dev_uninit);
1225 }
1226 
1227 static struct rte_pci_driver rte_cxgbe_pmd = {
1228 	.id_table = cxgb4_pci_tbl,
1229 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1230 	.probe = eth_cxgbe_pci_probe,
1231 	.remove = eth_cxgbe_pci_remove,
1232 };
1233 
1234 RTE_PMD_REGISTER_PCI(net_cxgbe, rte_cxgbe_pmd);
1235 RTE_PMD_REGISTER_PCI_TABLE(net_cxgbe, cxgb4_pci_tbl);
1236 RTE_PMD_REGISTER_KMOD_DEP(net_cxgbe, "* igb_uio | uio_pci_generic | vfio-pci");
1237 RTE_PMD_REGISTER_PARAM_STRING(net_cxgbe,
1238 			      CXGBE_DEVARG_KEEP_OVLAN "=<0|1> "
1239 			      CXGBE_DEVARG_FORCE_LINK_UP "=<0|1> ");
1240