xref: /dpdk/drivers/net/cxgbe/cxgbe_ethdev.c (revision 8809f78c7dd9f33a44a4f89c58fc91ded34296ed)
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 "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 	struct adapter *adapter = pi->adapter;
197 	struct sge *s = &adapter->sge;
198 	struct rte_eth_link new_link = { 0 };
199 	unsigned int i, work_done, budget = 32;
200 	u8 old_link = pi->link_cfg.link_ok;
201 
202 	for (i = 0; i < CXGBE_LINK_STATUS_POLL_CNT; i++) {
203 		if (!s->fw_evtq.desc)
204 			break;
205 
206 		cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
207 
208 		/* Exit if link status changed or always forced up */
209 		if (pi->link_cfg.link_ok != old_link ||
210 		    cxgbe_force_linkup(adapter))
211 			break;
212 
213 		if (!wait_to_complete)
214 			break;
215 
216 		rte_delay_ms(CXGBE_LINK_STATUS_POLL_MS);
217 	}
218 
219 	new_link.link_status = cxgbe_force_linkup(adapter) ?
220 			       ETH_LINK_UP : pi->link_cfg.link_ok;
221 	new_link.link_autoneg = pi->link_cfg.autoneg;
222 	new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
223 	new_link.link_speed = pi->link_cfg.speed;
224 
225 	return rte_eth_linkstatus_set(eth_dev, &new_link);
226 }
227 
228 /**
229  * Set device link up.
230  */
231 int cxgbe_dev_set_link_up(struct rte_eth_dev *dev)
232 {
233 	struct port_info *pi = dev->data->dev_private;
234 	struct adapter *adapter = pi->adapter;
235 	unsigned int work_done, budget = 32;
236 	struct sge *s = &adapter->sge;
237 	int ret;
238 
239 	if (!s->fw_evtq.desc)
240 		return -ENOMEM;
241 
242 	/* Flush all link events */
243 	cxgbe_poll(&s->fw_evtq, NULL, budget, &work_done);
244 
245 	/* If link already up, nothing to do */
246 	if (pi->link_cfg.link_ok)
247 		return 0;
248 
249 	ret = cxgbe_set_link_status(pi, true);
250 	if (ret)
251 		return ret;
252 
253 	cxgbe_dev_link_update(dev, 1);
254 	return 0;
255 }
256 
257 /**
258  * Set device link down.
259  */
260 int cxgbe_dev_set_link_down(struct rte_eth_dev *dev)
261 {
262 	struct port_info *pi = dev->data->dev_private;
263 	struct adapter *adapter = pi->adapter;
264 	unsigned int work_done, budget = 32;
265 	struct sge *s = &adapter->sge;
266 	int ret;
267 
268 	if (!s->fw_evtq.desc)
269 		return -ENOMEM;
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 = 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 + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
293 
294 	err = cxgbe_dev_info_get(eth_dev, &dev_info);
295 	if (err != 0)
296 		return err;
297 
298 	/* Must accommodate at least RTE_ETHER_MIN_MTU */
299 	if (new_mtu < RTE_ETHER_MIN_MTU || new_mtu > dev_info.max_rx_pktlen)
300 		return -EINVAL;
301 
302 	/* set to jumbo mode if needed */
303 	if (new_mtu > RTE_ETHER_MAX_LEN)
304 		eth_dev->data->dev_conf.rxmode.offloads |=
305 			DEV_RX_OFFLOAD_JUMBO_FRAME;
306 	else
307 		eth_dev->data->dev_conf.rxmode.offloads &=
308 			~DEV_RX_OFFLOAD_JUMBO_FRAME;
309 
310 	err = t4_set_rxmode(adapter, adapter->mbox, pi->viid, new_mtu, -1, -1,
311 			    -1, -1, true);
312 	if (!err)
313 		eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_mtu;
314 
315 	return err;
316 }
317 
318 /*
319  * Stop device.
320  */
321 int cxgbe_dev_close(struct rte_eth_dev *eth_dev)
322 {
323 	struct port_info *temp_pi, *pi = eth_dev->data->dev_private;
324 	struct adapter *adapter = pi->adapter;
325 	u8 i;
326 
327 	CXGBE_FUNC_TRACE();
328 
329 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
330 		return 0;
331 
332 	if (!(adapter->flags & FULL_INIT_DONE))
333 		return 0;
334 
335 	if (!pi->viid)
336 		return 0;
337 
338 	cxgbe_down(pi);
339 	t4_sge_eth_release_queues(pi);
340 	t4_free_vi(adapter, adapter->mbox, adapter->pf, 0, pi->viid);
341 	pi->viid = 0;
342 
343 	/* Free up the adapter-wide resources only after all the ports
344 	 * under this PF have been closed.
345 	 */
346 	for_each_port(adapter, i) {
347 		temp_pi = adap2pinfo(adapter, i);
348 		if (temp_pi->viid)
349 			return 0;
350 	}
351 
352 	cxgbe_close(adapter);
353 	rte_free(adapter);
354 
355 	return 0;
356 }
357 
358 /* Start the device.
359  * It returns 0 on success.
360  */
361 int cxgbe_dev_start(struct rte_eth_dev *eth_dev)
362 {
363 	struct port_info *pi = eth_dev->data->dev_private;
364 	struct rte_eth_rxmode *rx_conf = &eth_dev->data->dev_conf.rxmode;
365 	struct adapter *adapter = pi->adapter;
366 	int err = 0, i;
367 
368 	CXGBE_FUNC_TRACE();
369 
370 	/*
371 	 * If we don't have a connection to the firmware there's nothing we
372 	 * can do.
373 	 */
374 	if (!(adapter->flags & FW_OK)) {
375 		err = -ENXIO;
376 		goto out;
377 	}
378 
379 	if (!(adapter->flags & FULL_INIT_DONE)) {
380 		err = cxgbe_up(adapter);
381 		if (err < 0)
382 			goto out;
383 	}
384 
385 	if (rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER)
386 		eth_dev->data->scattered_rx = 1;
387 	else
388 		eth_dev->data->scattered_rx = 0;
389 
390 	cxgbe_enable_rx_queues(pi);
391 
392 	err = cxgbe_setup_rss(pi);
393 	if (err)
394 		goto out;
395 
396 	for (i = 0; i < pi->n_tx_qsets; i++) {
397 		err = cxgbe_dev_tx_queue_start(eth_dev, i);
398 		if (err)
399 			goto out;
400 	}
401 
402 	for (i = 0; i < pi->n_rx_qsets; i++) {
403 		err = cxgbe_dev_rx_queue_start(eth_dev, i);
404 		if (err)
405 			goto out;
406 	}
407 
408 	err = cxgbe_link_start(pi);
409 	if (err)
410 		goto out;
411 
412 out:
413 	return err;
414 }
415 
416 /*
417  * Stop device: disable rx and tx functions to allow for reconfiguring.
418  */
419 int cxgbe_dev_stop(struct rte_eth_dev *eth_dev)
420 {
421 	struct port_info *pi = eth_dev->data->dev_private;
422 	struct adapter *adapter = pi->adapter;
423 
424 	CXGBE_FUNC_TRACE();
425 
426 	if (!(adapter->flags & FULL_INIT_DONE))
427 		return 0;
428 
429 	cxgbe_down(pi);
430 
431 	/*
432 	 *  We clear queues only if both tx and rx path of the port
433 	 *  have been disabled
434 	 */
435 	t4_sge_eth_clear_queues(pi);
436 	eth_dev->data->scattered_rx = 0;
437 
438 	return 0;
439 }
440 
441 int cxgbe_dev_configure(struct rte_eth_dev *eth_dev)
442 {
443 	struct port_info *pi = eth_dev->data->dev_private;
444 	struct adapter *adapter = pi->adapter;
445 	int err;
446 
447 	CXGBE_FUNC_TRACE();
448 
449 	if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
450 		eth_dev->data->dev_conf.rxmode.offloads |=
451 			DEV_RX_OFFLOAD_RSS_HASH;
452 
453 	if (!(adapter->flags & FW_QUEUE_BOUND)) {
454 		err = cxgbe_setup_sge_fwevtq(adapter);
455 		if (err)
456 			return err;
457 		adapter->flags |= FW_QUEUE_BOUND;
458 		if (is_pf4(adapter)) {
459 			err = cxgbe_setup_sge_ctrl_txq(adapter);
460 			if (err)
461 				return err;
462 		}
463 	}
464 
465 	err = cxgbe_cfg_queue_count(eth_dev);
466 	if (err)
467 		return err;
468 
469 	return 0;
470 }
471 
472 int cxgbe_dev_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
473 {
474 	int ret;
475 	struct sge_eth_txq *txq = (struct sge_eth_txq *)
476 				  (eth_dev->data->tx_queues[tx_queue_id]);
477 
478 	dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
479 
480 	ret = t4_sge_eth_txq_start(txq);
481 	if (ret == 0)
482 		eth_dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
483 
484 	return ret;
485 }
486 
487 int cxgbe_dev_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
488 {
489 	int ret;
490 	struct sge_eth_txq *txq = (struct sge_eth_txq *)
491 				  (eth_dev->data->tx_queues[tx_queue_id]);
492 
493 	dev_debug(NULL, "%s: tx_queue_id = %d\n", __func__, tx_queue_id);
494 
495 	ret = t4_sge_eth_txq_stop(txq);
496 	if (ret == 0)
497 		eth_dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
498 
499 	return ret;
500 }
501 
502 int cxgbe_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
503 			     uint16_t queue_idx, uint16_t nb_desc,
504 			     unsigned int socket_id,
505 			     const struct rte_eth_txconf *tx_conf __rte_unused)
506 {
507 	struct port_info *pi = eth_dev->data->dev_private;
508 	struct adapter *adapter = pi->adapter;
509 	struct sge *s = &adapter->sge;
510 	unsigned int temp_nb_desc;
511 	struct sge_eth_txq *txq;
512 	int err = 0;
513 
514 	txq = &s->ethtxq[pi->first_txqset + queue_idx];
515 	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",
516 		  __func__, eth_dev->data->nb_tx_queues, queue_idx, nb_desc,
517 		  socket_id, pi->first_txqset);
518 
519 	/*  Free up the existing queue  */
520 	if (eth_dev->data->tx_queues[queue_idx]) {
521 		cxgbe_dev_tx_queue_release(eth_dev->data->tx_queues[queue_idx]);
522 		eth_dev->data->tx_queues[queue_idx] = NULL;
523 	}
524 
525 	eth_dev->data->tx_queues[queue_idx] = (void *)txq;
526 
527 	/* Sanity Checking
528 	 *
529 	 * nb_desc should be > 1023 and <= CXGBE_MAX_RING_DESC_SIZE
530 	 */
531 	temp_nb_desc = nb_desc;
532 	if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
533 		dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
534 			 __func__, CXGBE_MIN_RING_DESC_SIZE,
535 			 CXGBE_DEFAULT_TX_DESC_SIZE);
536 		temp_nb_desc = CXGBE_DEFAULT_TX_DESC_SIZE;
537 	} else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
538 		dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
539 			__func__, CXGBE_MIN_RING_DESC_SIZE,
540 			CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_TX_DESC_SIZE);
541 		return -(EINVAL);
542 	}
543 
544 	txq->q.size = temp_nb_desc;
545 
546 	err = t4_sge_alloc_eth_txq(adapter, txq, eth_dev, queue_idx,
547 				   s->fw_evtq.cntxt_id, socket_id);
548 
549 	dev_debug(adapter, "%s: txq->q.cntxt_id= %u txq->q.abs_id= %u err = %d\n",
550 		  __func__, txq->q.cntxt_id, txq->q.abs_id, err);
551 	return err;
552 }
553 
554 void cxgbe_dev_tx_queue_release(void *q)
555 {
556 	struct sge_eth_txq *txq = (struct sge_eth_txq *)q;
557 
558 	if (txq) {
559 		struct port_info *pi = (struct port_info *)
560 				       (txq->eth_dev->data->dev_private);
561 		struct adapter *adap = pi->adapter;
562 
563 		dev_debug(adapter, "%s: pi->port_id = %d; tx_queue_id = %d\n",
564 			  __func__, pi->port_id, txq->q.cntxt_id);
565 
566 		t4_sge_eth_txq_release(adap, txq);
567 	}
568 }
569 
570 int cxgbe_dev_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
571 {
572 	struct port_info *pi = eth_dev->data->dev_private;
573 	struct adapter *adap = pi->adapter;
574 	struct sge_eth_rxq *rxq;
575 	int ret;
576 
577 	dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
578 		  __func__, pi->port_id, rx_queue_id);
579 
580 	rxq = eth_dev->data->rx_queues[rx_queue_id];
581 	ret = t4_sge_eth_rxq_start(adap, rxq);
582 	if (ret == 0)
583 		eth_dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
584 
585 	return ret;
586 }
587 
588 int cxgbe_dev_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
589 {
590 	struct port_info *pi = eth_dev->data->dev_private;
591 	struct adapter *adap = pi->adapter;
592 	struct sge_eth_rxq *rxq;
593 	int ret;
594 
595 	dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
596 		  __func__, pi->port_id, rx_queue_id);
597 
598 	rxq = eth_dev->data->rx_queues[rx_queue_id];
599 	ret = t4_sge_eth_rxq_stop(adap, rxq);
600 	if (ret == 0)
601 		eth_dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
602 
603 	return ret;
604 }
605 
606 int cxgbe_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
607 			     uint16_t queue_idx, uint16_t nb_desc,
608 			     unsigned int socket_id,
609 			     const struct rte_eth_rxconf *rx_conf __rte_unused,
610 			     struct rte_mempool *mp)
611 {
612 	unsigned int pkt_len = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
613 	struct port_info *pi = eth_dev->data->dev_private;
614 	struct adapter *adapter = pi->adapter;
615 	struct rte_eth_dev_info dev_info;
616 	struct sge *s = &adapter->sge;
617 	unsigned int temp_nb_desc;
618 	int err = 0, msi_idx = 0;
619 	struct sge_eth_rxq *rxq;
620 
621 	rxq = &s->ethrxq[pi->first_rxqset + queue_idx];
622 	dev_debug(adapter, "%s: eth_dev->data->nb_rx_queues = %d; queue_idx = %d; nb_desc = %d; socket_id = %d; mp = %p\n",
623 		  __func__, eth_dev->data->nb_rx_queues, queue_idx, nb_desc,
624 		  socket_id, mp);
625 
626 	err = cxgbe_dev_info_get(eth_dev, &dev_info);
627 	if (err != 0) {
628 		dev_err(adap, "%s: error during getting ethernet device info",
629 			__func__);
630 		return err;
631 	}
632 
633 	/* Must accommodate at least RTE_ETHER_MIN_MTU */
634 	if ((pkt_len < dev_info.min_rx_bufsize) ||
635 	    (pkt_len > dev_info.max_rx_pktlen)) {
636 		dev_err(adap, "%s: max pkt len must be > %d and <= %d\n",
637 			__func__, dev_info.min_rx_bufsize,
638 			dev_info.max_rx_pktlen);
639 		return -EINVAL;
640 	}
641 
642 	/*  Free up the existing queue  */
643 	if (eth_dev->data->rx_queues[queue_idx]) {
644 		cxgbe_dev_rx_queue_release(eth_dev->data->rx_queues[queue_idx]);
645 		eth_dev->data->rx_queues[queue_idx] = NULL;
646 	}
647 
648 	eth_dev->data->rx_queues[queue_idx] = (void *)rxq;
649 
650 	/* Sanity Checking
651 	 *
652 	 * nb_desc should be > 0 and <= CXGBE_MAX_RING_DESC_SIZE
653 	 */
654 	temp_nb_desc = nb_desc;
655 	if (nb_desc < CXGBE_MIN_RING_DESC_SIZE) {
656 		dev_warn(adapter, "%s: number of descriptors must be >= %d. Using default [%d]\n",
657 			 __func__, CXGBE_MIN_RING_DESC_SIZE,
658 			 CXGBE_DEFAULT_RX_DESC_SIZE);
659 		temp_nb_desc = CXGBE_DEFAULT_RX_DESC_SIZE;
660 	} else if (nb_desc > CXGBE_MAX_RING_DESC_SIZE) {
661 		dev_err(adapter, "%s: number of descriptors must be between %d and %d inclusive. Default [%d]\n",
662 			__func__, CXGBE_MIN_RING_DESC_SIZE,
663 			CXGBE_MAX_RING_DESC_SIZE, CXGBE_DEFAULT_RX_DESC_SIZE);
664 		return -(EINVAL);
665 	}
666 
667 	rxq->rspq.size = temp_nb_desc;
668 	if ((&rxq->fl) != NULL)
669 		rxq->fl.size = temp_nb_desc;
670 
671 	/* Set to jumbo mode if necessary */
672 	if (pkt_len > RTE_ETHER_MAX_LEN)
673 		eth_dev->data->dev_conf.rxmode.offloads |=
674 			DEV_RX_OFFLOAD_JUMBO_FRAME;
675 	else
676 		eth_dev->data->dev_conf.rxmode.offloads &=
677 			~DEV_RX_OFFLOAD_JUMBO_FRAME;
678 
679 	err = t4_sge_alloc_rxq(adapter, &rxq->rspq, false, eth_dev, msi_idx,
680 			       &rxq->fl, NULL,
681 			       is_pf4(adapter) ?
682 			       t4_get_tp_ch_map(adapter, pi->tx_chan) : 0, mp,
683 			       queue_idx, socket_id);
684 
685 	dev_debug(adapter, "%s: err = %d; port_id = %d; cntxt_id = %u; abs_id = %u\n",
686 		  __func__, err, pi->port_id, rxq->rspq.cntxt_id,
687 		  rxq->rspq.abs_id);
688 	return err;
689 }
690 
691 void cxgbe_dev_rx_queue_release(void *q)
692 {
693 	struct sge_eth_rxq *rxq = (struct sge_eth_rxq *)q;
694 
695 	if (rxq) {
696 		struct port_info *pi = (struct port_info *)
697 				       (rxq->rspq.eth_dev->data->dev_private);
698 		struct adapter *adap = pi->adapter;
699 
700 		dev_debug(adapter, "%s: pi->port_id = %d; rx_queue_id = %d\n",
701 			  __func__, pi->port_id, rxq->rspq.cntxt_id);
702 
703 		t4_sge_eth_rxq_release(adap, rxq);
704 	}
705 }
706 
707 /*
708  * Get port statistics.
709  */
710 static int cxgbe_dev_stats_get(struct rte_eth_dev *eth_dev,
711 				struct rte_eth_stats *eth_stats)
712 {
713 	struct port_info *pi = eth_dev->data->dev_private;
714 	struct adapter *adapter = pi->adapter;
715 	struct sge *s = &adapter->sge;
716 	struct port_stats ps;
717 	unsigned int i;
718 
719 	cxgbe_stats_get(pi, &ps);
720 
721 	/* RX Stats */
722 	eth_stats->imissed  = ps.rx_ovflow0 + ps.rx_ovflow1 +
723 			      ps.rx_ovflow2 + ps.rx_ovflow3 +
724 			      ps.rx_trunc0 + ps.rx_trunc1 +
725 			      ps.rx_trunc2 + ps.rx_trunc3;
726 	eth_stats->ierrors  = ps.rx_symbol_err + ps.rx_fcs_err +
727 			      ps.rx_jabber + ps.rx_too_long + ps.rx_runt +
728 			      ps.rx_len_err;
729 
730 	/* TX Stats */
731 	eth_stats->opackets = ps.tx_frames;
732 	eth_stats->obytes   = ps.tx_octets;
733 	eth_stats->oerrors  = ps.tx_error_frames;
734 
735 	for (i = 0; i < pi->n_rx_qsets; i++) {
736 		struct sge_eth_rxq *rxq =
737 			&s->ethrxq[pi->first_rxqset + i];
738 
739 		eth_stats->q_ipackets[i] = rxq->stats.pkts;
740 		eth_stats->q_ibytes[i] = rxq->stats.rx_bytes;
741 		eth_stats->ipackets += eth_stats->q_ipackets[i];
742 		eth_stats->ibytes += eth_stats->q_ibytes[i];
743 	}
744 
745 	for (i = 0; i < pi->n_tx_qsets; i++) {
746 		struct sge_eth_txq *txq =
747 			&s->ethtxq[pi->first_txqset + i];
748 
749 		eth_stats->q_opackets[i] = txq->stats.pkts;
750 		eth_stats->q_obytes[i] = txq->stats.tx_bytes;
751 	}
752 	return 0;
753 }
754 
755 /*
756  * Reset port statistics.
757  */
758 static int cxgbe_dev_stats_reset(struct rte_eth_dev *eth_dev)
759 {
760 	struct port_info *pi = eth_dev->data->dev_private;
761 	struct adapter *adapter = pi->adapter;
762 	struct sge *s = &adapter->sge;
763 	unsigned int i;
764 
765 	cxgbe_stats_reset(pi);
766 	for (i = 0; i < pi->n_rx_qsets; i++) {
767 		struct sge_eth_rxq *rxq =
768 			&s->ethrxq[pi->first_rxqset + i];
769 
770 		rxq->stats.pkts = 0;
771 		rxq->stats.rx_bytes = 0;
772 	}
773 	for (i = 0; i < pi->n_tx_qsets; i++) {
774 		struct sge_eth_txq *txq =
775 			&s->ethtxq[pi->first_txqset + i];
776 
777 		txq->stats.pkts = 0;
778 		txq->stats.tx_bytes = 0;
779 		txq->stats.mapping_err = 0;
780 	}
781 
782 	return 0;
783 }
784 
785 static int cxgbe_flow_ctrl_get(struct rte_eth_dev *eth_dev,
786 			       struct rte_eth_fc_conf *fc_conf)
787 {
788 	struct port_info *pi = eth_dev->data->dev_private;
789 	struct link_config *lc = &pi->link_cfg;
790 	int rx_pause, tx_pause;
791 
792 	fc_conf->autoneg = lc->fc & PAUSE_AUTONEG;
793 	rx_pause = lc->fc & PAUSE_RX;
794 	tx_pause = lc->fc & PAUSE_TX;
795 
796 	if (rx_pause && tx_pause)
797 		fc_conf->mode = RTE_FC_FULL;
798 	else if (rx_pause)
799 		fc_conf->mode = RTE_FC_RX_PAUSE;
800 	else if (tx_pause)
801 		fc_conf->mode = RTE_FC_TX_PAUSE;
802 	else
803 		fc_conf->mode = RTE_FC_NONE;
804 	return 0;
805 }
806 
807 static int cxgbe_flow_ctrl_set(struct rte_eth_dev *eth_dev,
808 			       struct rte_eth_fc_conf *fc_conf)
809 {
810 	struct port_info *pi = eth_dev->data->dev_private;
811 	struct adapter *adapter = pi->adapter;
812 	struct link_config *lc = &pi->link_cfg;
813 
814 	if (lc->pcaps & FW_PORT_CAP32_ANEG) {
815 		if (fc_conf->autoneg)
816 			lc->requested_fc |= PAUSE_AUTONEG;
817 		else
818 			lc->requested_fc &= ~PAUSE_AUTONEG;
819 	}
820 
821 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
822 	    (fc_conf->mode & RTE_FC_RX_PAUSE))
823 		lc->requested_fc |= PAUSE_RX;
824 	else
825 		lc->requested_fc &= ~PAUSE_RX;
826 
827 	if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
828 	    (fc_conf->mode & RTE_FC_TX_PAUSE))
829 		lc->requested_fc |= PAUSE_TX;
830 	else
831 		lc->requested_fc &= ~PAUSE_TX;
832 
833 	return t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
834 			     &pi->link_cfg);
835 }
836 
837 const uint32_t *
838 cxgbe_dev_supported_ptypes_get(struct rte_eth_dev *eth_dev)
839 {
840 	static const uint32_t ptypes[] = {
841 		RTE_PTYPE_L3_IPV4,
842 		RTE_PTYPE_L3_IPV6,
843 		RTE_PTYPE_UNKNOWN
844 	};
845 
846 	if (eth_dev->rx_pkt_burst == cxgbe_recv_pkts)
847 		return ptypes;
848 	return NULL;
849 }
850 
851 /* Update RSS hash configuration
852  */
853 static int cxgbe_dev_rss_hash_update(struct rte_eth_dev *dev,
854 				     struct rte_eth_rss_conf *rss_conf)
855 {
856 	struct port_info *pi = dev->data->dev_private;
857 	struct adapter *adapter = pi->adapter;
858 	int err;
859 
860 	err = cxgbe_write_rss_conf(pi, rss_conf->rss_hf);
861 	if (err)
862 		return err;
863 
864 	pi->rss_hf = rss_conf->rss_hf;
865 
866 	if (rss_conf->rss_key) {
867 		u32 key[10], mod_key[10];
868 		int i, j;
869 
870 		memcpy(key, rss_conf->rss_key, CXGBE_DEFAULT_RSS_KEY_LEN);
871 
872 		for (i = 9, j = 0; i >= 0; i--, j++)
873 			mod_key[j] = cpu_to_be32(key[i]);
874 
875 		t4_write_rss_key(adapter, mod_key, -1);
876 	}
877 
878 	return 0;
879 }
880 
881 /* Get RSS hash configuration
882  */
883 static int cxgbe_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
884 				       struct rte_eth_rss_conf *rss_conf)
885 {
886 	struct port_info *pi = dev->data->dev_private;
887 	struct adapter *adapter = pi->adapter;
888 	u64 rss_hf = 0;
889 	u64 flags = 0;
890 	int err;
891 
892 	err = t4_read_config_vi_rss(adapter, adapter->mbox, pi->viid,
893 				    &flags, NULL);
894 
895 	if (err)
896 		return err;
897 
898 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) {
899 		rss_hf |= CXGBE_RSS_HF_TCP_IPV6_MASK;
900 		if (flags & F_FW_RSS_VI_CONFIG_CMD_UDPEN)
901 			rss_hf |= CXGBE_RSS_HF_UDP_IPV6_MASK;
902 	}
903 
904 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
905 		rss_hf |= CXGBE_RSS_HF_IPV6_MASK;
906 
907 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) {
908 		rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
909 		if (flags & F_FW_RSS_VI_CONFIG_CMD_UDPEN)
910 			rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
911 	}
912 
913 	if (flags & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
914 		rss_hf |= CXGBE_RSS_HF_IPV4_MASK;
915 
916 	rss_conf->rss_hf = rss_hf;
917 
918 	if (rss_conf->rss_key) {
919 		u32 key[10], mod_key[10];
920 		int i, j;
921 
922 		t4_read_rss_key(adapter, key);
923 
924 		for (i = 9, j = 0; i >= 0; i--, j++)
925 			mod_key[j] = be32_to_cpu(key[i]);
926 
927 		memcpy(rss_conf->rss_key, mod_key, CXGBE_DEFAULT_RSS_KEY_LEN);
928 	}
929 
930 	return 0;
931 }
932 
933 static int cxgbe_dev_rss_reta_update(struct rte_eth_dev *dev,
934 				     struct rte_eth_rss_reta_entry64 *reta_conf,
935 				     uint16_t reta_size)
936 {
937 	struct port_info *pi = dev->data->dev_private;
938 	struct adapter *adapter = pi->adapter;
939 	u16 i, idx, shift, *rss;
940 	int ret;
941 
942 	if (!(adapter->flags & FULL_INIT_DONE))
943 		return -ENOMEM;
944 
945 	if (!reta_size || reta_size > pi->rss_size)
946 		return -EINVAL;
947 
948 	rss = rte_calloc(NULL, pi->rss_size, sizeof(u16), 0);
949 	if (!rss)
950 		return -ENOMEM;
951 
952 	rte_memcpy(rss, pi->rss, pi->rss_size * sizeof(u16));
953 	for (i = 0; i < reta_size; i++) {
954 		idx = i / RTE_RETA_GROUP_SIZE;
955 		shift = i % RTE_RETA_GROUP_SIZE;
956 		if (!(reta_conf[idx].mask & (1ULL << shift)))
957 			continue;
958 
959 		rss[i] = reta_conf[idx].reta[shift];
960 	}
961 
962 	ret = cxgbe_write_rss(pi, rss);
963 	if (!ret)
964 		rte_memcpy(pi->rss, rss, pi->rss_size * sizeof(u16));
965 
966 	rte_free(rss);
967 	return ret;
968 }
969 
970 static int cxgbe_dev_rss_reta_query(struct rte_eth_dev *dev,
971 				    struct rte_eth_rss_reta_entry64 *reta_conf,
972 				    uint16_t reta_size)
973 {
974 	struct port_info *pi = dev->data->dev_private;
975 	struct adapter *adapter = pi->adapter;
976 	u16 i, idx, shift;
977 
978 	if (!(adapter->flags & FULL_INIT_DONE))
979 		return -ENOMEM;
980 
981 	if (!reta_size || reta_size > pi->rss_size)
982 		return -EINVAL;
983 
984 	for (i = 0; i < reta_size; i++) {
985 		idx = i / RTE_RETA_GROUP_SIZE;
986 		shift = i % RTE_RETA_GROUP_SIZE;
987 		if (!(reta_conf[idx].mask & (1ULL << shift)))
988 			continue;
989 
990 		reta_conf[idx].reta[shift] = pi->rss[i];
991 	}
992 
993 	return 0;
994 }
995 
996 static int cxgbe_get_eeprom_length(struct rte_eth_dev *dev)
997 {
998 	RTE_SET_USED(dev);
999 	return EEPROMSIZE;
1000 }
1001 
1002 /**
1003  * eeprom_ptov - translate a physical EEPROM address to virtual
1004  * @phys_addr: the physical EEPROM address
1005  * @fn: the PCI function number
1006  * @sz: size of function-specific area
1007  *
1008  * Translate a physical EEPROM address to virtual.  The first 1K is
1009  * accessed through virtual addresses starting at 31K, the rest is
1010  * accessed through virtual addresses starting at 0.
1011  *
1012  * The mapping is as follows:
1013  * [0..1K) -> [31K..32K)
1014  * [1K..1K+A) -> [31K-A..31K)
1015  * [1K+A..ES) -> [0..ES-A-1K)
1016  *
1017  * where A = @fn * @sz, and ES = EEPROM size.
1018  */
1019 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
1020 {
1021 	fn *= sz;
1022 	if (phys_addr < 1024)
1023 		return phys_addr + (31 << 10);
1024 	if (phys_addr < 1024 + fn)
1025 		return fn + phys_addr - 1024;
1026 	if (phys_addr < EEPROMSIZE)
1027 		return phys_addr - 1024 - fn;
1028 	if (phys_addr < EEPROMVSIZE)
1029 		return phys_addr - 1024;
1030 	return -EINVAL;
1031 }
1032 
1033 /* The next two routines implement eeprom read/write from physical addresses.
1034  */
1035 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
1036 {
1037 	int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
1038 
1039 	if (vaddr >= 0)
1040 		vaddr = t4_seeprom_read(adap, vaddr, v);
1041 	return vaddr < 0 ? vaddr : 0;
1042 }
1043 
1044 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
1045 {
1046 	int vaddr = eeprom_ptov(phys_addr, adap->pf, EEPROMPFSIZE);
1047 
1048 	if (vaddr >= 0)
1049 		vaddr = t4_seeprom_write(adap, vaddr, v);
1050 	return vaddr < 0 ? vaddr : 0;
1051 }
1052 
1053 #define EEPROM_MAGIC 0x38E2F10C
1054 
1055 static int cxgbe_get_eeprom(struct rte_eth_dev *dev,
1056 			    struct rte_dev_eeprom_info *e)
1057 {
1058 	struct port_info *pi = dev->data->dev_private;
1059 	struct adapter *adapter = pi->adapter;
1060 	u32 i, err = 0;
1061 	u8 *buf = rte_zmalloc(NULL, EEPROMSIZE, 0);
1062 
1063 	if (!buf)
1064 		return -ENOMEM;
1065 
1066 	e->magic = EEPROM_MAGIC;
1067 	for (i = e->offset & ~3; !err && i < e->offset + e->length; i += 4)
1068 		err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
1069 
1070 	if (!err)
1071 		rte_memcpy(e->data, buf + e->offset, e->length);
1072 	rte_free(buf);
1073 	return err;
1074 }
1075 
1076 static int cxgbe_set_eeprom(struct rte_eth_dev *dev,
1077 			    struct rte_dev_eeprom_info *eeprom)
1078 {
1079 	struct port_info *pi = dev->data->dev_private;
1080 	struct adapter *adapter = pi->adapter;
1081 	u8 *buf;
1082 	int err = 0;
1083 	u32 aligned_offset, aligned_len, *p;
1084 
1085 	if (eeprom->magic != EEPROM_MAGIC)
1086 		return -EINVAL;
1087 
1088 	aligned_offset = eeprom->offset & ~3;
1089 	aligned_len = (eeprom->length + (eeprom->offset & 3) + 3) & ~3;
1090 
1091 	if (adapter->pf > 0) {
1092 		u32 start = 1024 + adapter->pf * EEPROMPFSIZE;
1093 
1094 		if (aligned_offset < start ||
1095 		    aligned_offset + aligned_len > start + EEPROMPFSIZE)
1096 			return -EPERM;
1097 	}
1098 
1099 	if (aligned_offset != eeprom->offset || aligned_len != eeprom->length) {
1100 		/* RMW possibly needed for first or last words.
1101 		 */
1102 		buf = rte_zmalloc(NULL, aligned_len, 0);
1103 		if (!buf)
1104 			return -ENOMEM;
1105 		err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
1106 		if (!err && aligned_len > 4)
1107 			err = eeprom_rd_phys(adapter,
1108 					     aligned_offset + aligned_len - 4,
1109 					     (u32 *)&buf[aligned_len - 4]);
1110 		if (err)
1111 			goto out;
1112 		rte_memcpy(buf + (eeprom->offset & 3), eeprom->data,
1113 			   eeprom->length);
1114 	} else {
1115 		buf = eeprom->data;
1116 	}
1117 
1118 	err = t4_seeprom_wp(adapter, false);
1119 	if (err)
1120 		goto out;
1121 
1122 	for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
1123 		err = eeprom_wr_phys(adapter, aligned_offset, *p);
1124 		aligned_offset += 4;
1125 	}
1126 
1127 	if (!err)
1128 		err = t4_seeprom_wp(adapter, true);
1129 out:
1130 	if (buf != eeprom->data)
1131 		rte_free(buf);
1132 	return err;
1133 }
1134 
1135 static int cxgbe_get_regs_len(struct rte_eth_dev *eth_dev)
1136 {
1137 	struct port_info *pi = eth_dev->data->dev_private;
1138 	struct adapter *adapter = pi->adapter;
1139 
1140 	return t4_get_regs_len(adapter) / sizeof(uint32_t);
1141 }
1142 
1143 static int cxgbe_get_regs(struct rte_eth_dev *eth_dev,
1144 			  struct rte_dev_reg_info *regs)
1145 {
1146 	struct port_info *pi = eth_dev->data->dev_private;
1147 	struct adapter *adapter = pi->adapter;
1148 
1149 	regs->version = CHELSIO_CHIP_VERSION(adapter->params.chip) |
1150 		(CHELSIO_CHIP_RELEASE(adapter->params.chip) << 10) |
1151 		(1 << 16);
1152 
1153 	if (regs->data == NULL) {
1154 		regs->length = cxgbe_get_regs_len(eth_dev);
1155 		regs->width = sizeof(uint32_t);
1156 
1157 		return 0;
1158 	}
1159 
1160 	t4_get_regs(adapter, regs->data, (regs->length * sizeof(uint32_t)));
1161 
1162 	return 0;
1163 }
1164 
1165 int cxgbe_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
1166 {
1167 	struct port_info *pi = dev->data->dev_private;
1168 	int ret;
1169 
1170 	ret = cxgbe_mpstcam_modify(pi, (int)pi->xact_addr_filt, (u8 *)addr);
1171 	if (ret < 0) {
1172 		dev_err(adapter, "failed to set mac addr; err = %d\n",
1173 			ret);
1174 		return ret;
1175 	}
1176 	pi->xact_addr_filt = ret;
1177 	return 0;
1178 }
1179 
1180 static const struct eth_dev_ops cxgbe_eth_dev_ops = {
1181 	.dev_start		= cxgbe_dev_start,
1182 	.dev_stop		= cxgbe_dev_stop,
1183 	.dev_close		= cxgbe_dev_close,
1184 	.promiscuous_enable	= cxgbe_dev_promiscuous_enable,
1185 	.promiscuous_disable	= cxgbe_dev_promiscuous_disable,
1186 	.allmulticast_enable	= cxgbe_dev_allmulticast_enable,
1187 	.allmulticast_disable	= cxgbe_dev_allmulticast_disable,
1188 	.dev_configure		= cxgbe_dev_configure,
1189 	.dev_infos_get		= cxgbe_dev_info_get,
1190 	.dev_supported_ptypes_get = cxgbe_dev_supported_ptypes_get,
1191 	.link_update		= cxgbe_dev_link_update,
1192 	.dev_set_link_up        = cxgbe_dev_set_link_up,
1193 	.dev_set_link_down      = cxgbe_dev_set_link_down,
1194 	.mtu_set		= cxgbe_dev_mtu_set,
1195 	.tx_queue_setup         = cxgbe_dev_tx_queue_setup,
1196 	.tx_queue_start		= cxgbe_dev_tx_queue_start,
1197 	.tx_queue_stop		= cxgbe_dev_tx_queue_stop,
1198 	.tx_queue_release	= cxgbe_dev_tx_queue_release,
1199 	.rx_queue_setup         = cxgbe_dev_rx_queue_setup,
1200 	.rx_queue_start		= cxgbe_dev_rx_queue_start,
1201 	.rx_queue_stop		= cxgbe_dev_rx_queue_stop,
1202 	.rx_queue_release	= cxgbe_dev_rx_queue_release,
1203 	.filter_ctrl            = cxgbe_dev_filter_ctrl,
1204 	.stats_get		= cxgbe_dev_stats_get,
1205 	.stats_reset		= cxgbe_dev_stats_reset,
1206 	.flow_ctrl_get		= cxgbe_flow_ctrl_get,
1207 	.flow_ctrl_set		= cxgbe_flow_ctrl_set,
1208 	.get_eeprom_length	= cxgbe_get_eeprom_length,
1209 	.get_eeprom		= cxgbe_get_eeprom,
1210 	.set_eeprom		= cxgbe_set_eeprom,
1211 	.get_reg		= cxgbe_get_regs,
1212 	.rss_hash_update	= cxgbe_dev_rss_hash_update,
1213 	.rss_hash_conf_get	= cxgbe_dev_rss_hash_conf_get,
1214 	.mac_addr_set		= cxgbe_mac_addr_set,
1215 	.reta_update            = cxgbe_dev_rss_reta_update,
1216 	.reta_query             = cxgbe_dev_rss_reta_query,
1217 };
1218 
1219 /*
1220  * Initialize driver
1221  * It returns 0 on success.
1222  */
1223 static int eth_cxgbe_dev_init(struct rte_eth_dev *eth_dev)
1224 {
1225 	struct rte_pci_device *pci_dev;
1226 	struct port_info *pi = eth_dev->data->dev_private;
1227 	struct adapter *adapter = NULL;
1228 	char name[RTE_ETH_NAME_MAX_LEN];
1229 	int err = 0;
1230 
1231 	CXGBE_FUNC_TRACE();
1232 
1233 	eth_dev->dev_ops = &cxgbe_eth_dev_ops;
1234 	eth_dev->rx_pkt_burst = &cxgbe_recv_pkts;
1235 	eth_dev->tx_pkt_burst = &cxgbe_xmit_pkts;
1236 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1237 
1238 	/* for secondary processes, we attach to ethdevs allocated by primary
1239 	 * and do minimal initialization.
1240 	 */
1241 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1242 		int i;
1243 
1244 		for (i = 1; i < MAX_NPORTS; i++) {
1245 			struct rte_eth_dev *rest_eth_dev;
1246 			char namei[RTE_ETH_NAME_MAX_LEN];
1247 
1248 			snprintf(namei, sizeof(namei), "%s_%d",
1249 				 pci_dev->device.name, i);
1250 			rest_eth_dev = rte_eth_dev_attach_secondary(namei);
1251 			if (rest_eth_dev) {
1252 				rest_eth_dev->device = &pci_dev->device;
1253 				rest_eth_dev->dev_ops =
1254 					eth_dev->dev_ops;
1255 				rest_eth_dev->rx_pkt_burst =
1256 					eth_dev->rx_pkt_burst;
1257 				rest_eth_dev->tx_pkt_burst =
1258 					eth_dev->tx_pkt_burst;
1259 				rte_eth_dev_probing_finish(rest_eth_dev);
1260 			}
1261 		}
1262 		return 0;
1263 	}
1264 
1265 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1266 
1267 	snprintf(name, sizeof(name), "cxgbeadapter%d", eth_dev->data->port_id);
1268 	adapter = rte_zmalloc(name, sizeof(*adapter), 0);
1269 	if (!adapter)
1270 		return -1;
1271 
1272 	adapter->use_unpacked_mode = 1;
1273 	adapter->regs = (void *)pci_dev->mem_resource[0].addr;
1274 	if (!adapter->regs) {
1275 		dev_err(adapter, "%s: cannot map device registers\n", __func__);
1276 		err = -ENOMEM;
1277 		goto out_free_adapter;
1278 	}
1279 	adapter->pdev = pci_dev;
1280 	adapter->eth_dev = eth_dev;
1281 	pi->adapter = adapter;
1282 
1283 	cxgbe_process_devargs(adapter);
1284 
1285 	err = cxgbe_probe(adapter);
1286 	if (err) {
1287 		dev_err(adapter, "%s: cxgbe probe failed with err %d\n",
1288 			__func__, err);
1289 		goto out_free_adapter;
1290 	}
1291 
1292 	return 0;
1293 
1294 out_free_adapter:
1295 	rte_free(adapter);
1296 	return err;
1297 }
1298 
1299 static int eth_cxgbe_dev_uninit(struct rte_eth_dev *eth_dev)
1300 {
1301 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1302 	uint16_t port_id;
1303 	int err = 0;
1304 
1305 	/* Free up other ports and all resources */
1306 	RTE_ETH_FOREACH_DEV_OF(port_id, &pci_dev->device)
1307 		err |= rte_eth_dev_close(port_id);
1308 
1309 	return err == 0 ? 0 : -EIO;
1310 }
1311 
1312 static int eth_cxgbe_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1313 	struct rte_pci_device *pci_dev)
1314 {
1315 	return rte_eth_dev_pci_generic_probe(pci_dev,
1316 		sizeof(struct port_info), eth_cxgbe_dev_init);
1317 }
1318 
1319 static int eth_cxgbe_pci_remove(struct rte_pci_device *pci_dev)
1320 {
1321 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_cxgbe_dev_uninit);
1322 }
1323 
1324 static struct rte_pci_driver rte_cxgbe_pmd = {
1325 	.id_table = cxgb4_pci_tbl,
1326 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1327 	.probe = eth_cxgbe_pci_probe,
1328 	.remove = eth_cxgbe_pci_remove,
1329 };
1330 
1331 RTE_PMD_REGISTER_PCI(net_cxgbe, rte_cxgbe_pmd);
1332 RTE_PMD_REGISTER_PCI_TABLE(net_cxgbe, cxgb4_pci_tbl);
1333 RTE_PMD_REGISTER_KMOD_DEP(net_cxgbe, "* igb_uio | uio_pci_generic | vfio-pci");
1334 RTE_PMD_REGISTER_PARAM_STRING(net_cxgbe,
1335 			      CXGBE_DEVARG_CMN_KEEP_OVLAN "=<0|1> "
1336 			      CXGBE_DEVARG_CMN_TX_MODE_LATENCY "=<0|1> "
1337 			      CXGBE_DEVARG_PF_FILTER_MODE "=<uint32> "
1338 			      CXGBE_DEVARG_PF_FILTER_MASK "=<uint32> ");
1339 RTE_LOG_REGISTER(cxgbe_logtype, pmd.net.cxgbe, NOTICE);
1340 RTE_LOG_REGISTER(cxgbe_mbox_logtype, pmd.net.cxgbe.mbox, NOTICE);
1341