1 /*
2  * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 
34 #include <linux/etherdevice.h>
35 #include <linux/tcp.h>
36 #include <linux/if_vlan.h>
37 #include <linux/delay.h>
38 #include <linux/slab.h>
39 #include <linux/hash.h>
40 #include <net/ip.h>
41 #include <net/busy_poll.h>
42 
43 #include <linux/mlx4/driver.h>
44 #include <linux/mlx4/device.h>
45 #include <linux/mlx4/cmd.h>
46 #include <linux/mlx4/cq.h>
47 
48 #include "mlx4_en.h"
49 #include "en_port.h"
50 
51 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
52 {
53 	struct mlx4_en_priv *priv = netdev_priv(dev);
54 	int i;
55 	unsigned int offset = 0;
56 
57 	if (up && up != MLX4_EN_NUM_UP)
58 		return -EINVAL;
59 
60 	netdev_set_num_tc(dev, up);
61 
62 	/* Partition Tx queues evenly amongst UP's */
63 	for (i = 0; i < up; i++) {
64 		netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
65 		offset += priv->num_tx_rings_p_up;
66 	}
67 
68 	return 0;
69 }
70 
71 #ifdef CONFIG_NET_RX_BUSY_POLL
72 /* must be called with local_bh_disable()d */
73 static int mlx4_en_low_latency_recv(struct napi_struct *napi)
74 {
75 	struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
76 	struct net_device *dev = cq->dev;
77 	struct mlx4_en_priv *priv = netdev_priv(dev);
78 	struct mlx4_en_rx_ring *rx_ring = priv->rx_ring[cq->ring];
79 	int done;
80 
81 	if (!priv->port_up)
82 		return LL_FLUSH_FAILED;
83 
84 	if (!mlx4_en_cq_lock_poll(cq))
85 		return LL_FLUSH_BUSY;
86 
87 	done = mlx4_en_process_rx_cq(dev, cq, 4);
88 	if (likely(done))
89 		rx_ring->cleaned += done;
90 	else
91 		rx_ring->misses++;
92 
93 	mlx4_en_cq_unlock_poll(cq);
94 
95 	return done;
96 }
97 #endif	/* CONFIG_NET_RX_BUSY_POLL */
98 
99 #ifdef CONFIG_RFS_ACCEL
100 
101 struct mlx4_en_filter {
102 	struct list_head next;
103 	struct work_struct work;
104 
105 	u8     ip_proto;
106 	__be32 src_ip;
107 	__be32 dst_ip;
108 	__be16 src_port;
109 	__be16 dst_port;
110 
111 	int rxq_index;
112 	struct mlx4_en_priv *priv;
113 	u32 flow_id;			/* RFS infrastructure id */
114 	int id;				/* mlx4_en driver id */
115 	u64 reg_id;			/* Flow steering API id */
116 	u8 activated;			/* Used to prevent expiry before filter
117 					 * is attached
118 					 */
119 	struct hlist_node filter_chain;
120 };
121 
122 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
123 
124 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
125 {
126 	switch (ip_proto) {
127 	case IPPROTO_UDP:
128 		return MLX4_NET_TRANS_RULE_ID_UDP;
129 	case IPPROTO_TCP:
130 		return MLX4_NET_TRANS_RULE_ID_TCP;
131 	default:
132 		return -EPROTONOSUPPORT;
133 	}
134 };
135 
136 static void mlx4_en_filter_work(struct work_struct *work)
137 {
138 	struct mlx4_en_filter *filter = container_of(work,
139 						     struct mlx4_en_filter,
140 						     work);
141 	struct mlx4_en_priv *priv = filter->priv;
142 	struct mlx4_spec_list spec_tcp_udp = {
143 		.id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
144 		{
145 			.tcp_udp = {
146 				.dst_port = filter->dst_port,
147 				.dst_port_msk = (__force __be16)-1,
148 				.src_port = filter->src_port,
149 				.src_port_msk = (__force __be16)-1,
150 			},
151 		},
152 	};
153 	struct mlx4_spec_list spec_ip = {
154 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
155 		{
156 			.ipv4 = {
157 				.dst_ip = filter->dst_ip,
158 				.dst_ip_msk = (__force __be32)-1,
159 				.src_ip = filter->src_ip,
160 				.src_ip_msk = (__force __be32)-1,
161 			},
162 		},
163 	};
164 	struct mlx4_spec_list spec_eth = {
165 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
166 	};
167 	struct mlx4_net_trans_rule rule = {
168 		.list = LIST_HEAD_INIT(rule.list),
169 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
170 		.exclusive = 1,
171 		.allow_loopback = 1,
172 		.promisc_mode = MLX4_FS_REGULAR,
173 		.port = priv->port,
174 		.priority = MLX4_DOMAIN_RFS,
175 	};
176 	int rc;
177 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
178 
179 	if (spec_tcp_udp.id < 0) {
180 		en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
181 			filter->ip_proto);
182 		goto ignore;
183 	}
184 	list_add_tail(&spec_eth.list, &rule.list);
185 	list_add_tail(&spec_ip.list, &rule.list);
186 	list_add_tail(&spec_tcp_udp.list, &rule.list);
187 
188 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
189 	memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
190 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
191 
192 	filter->activated = 0;
193 
194 	if (filter->reg_id) {
195 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
196 		if (rc && rc != -ENOENT)
197 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
198 	}
199 
200 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
201 	if (rc)
202 		en_err(priv, "Error attaching flow. err = %d\n", rc);
203 
204 ignore:
205 	mlx4_en_filter_rfs_expire(priv);
206 
207 	filter->activated = 1;
208 }
209 
210 static inline struct hlist_head *
211 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
212 		   __be16 src_port, __be16 dst_port)
213 {
214 	unsigned long l;
215 	int bucket_idx;
216 
217 	l = (__force unsigned long)src_port |
218 	    ((__force unsigned long)dst_port << 2);
219 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
220 
221 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
222 
223 	return &priv->filter_hash[bucket_idx];
224 }
225 
226 static struct mlx4_en_filter *
227 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
228 		     __be32 dst_ip, u8 ip_proto, __be16 src_port,
229 		     __be16 dst_port, u32 flow_id)
230 {
231 	struct mlx4_en_filter *filter = NULL;
232 
233 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
234 	if (!filter)
235 		return NULL;
236 
237 	filter->priv = priv;
238 	filter->rxq_index = rxq_index;
239 	INIT_WORK(&filter->work, mlx4_en_filter_work);
240 
241 	filter->src_ip = src_ip;
242 	filter->dst_ip = dst_ip;
243 	filter->ip_proto = ip_proto;
244 	filter->src_port = src_port;
245 	filter->dst_port = dst_port;
246 
247 	filter->flow_id = flow_id;
248 
249 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
250 
251 	list_add_tail(&filter->next, &priv->filters);
252 	hlist_add_head(&filter->filter_chain,
253 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
254 					  dst_port));
255 
256 	return filter;
257 }
258 
259 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
260 {
261 	struct mlx4_en_priv *priv = filter->priv;
262 	int rc;
263 
264 	list_del(&filter->next);
265 
266 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
267 	if (rc && rc != -ENOENT)
268 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
269 
270 	kfree(filter);
271 }
272 
273 static inline struct mlx4_en_filter *
274 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
275 		    u8 ip_proto, __be16 src_port, __be16 dst_port)
276 {
277 	struct mlx4_en_filter *filter;
278 	struct mlx4_en_filter *ret = NULL;
279 
280 	hlist_for_each_entry(filter,
281 			     filter_hash_bucket(priv, src_ip, dst_ip,
282 						src_port, dst_port),
283 			     filter_chain) {
284 		if (filter->src_ip == src_ip &&
285 		    filter->dst_ip == dst_ip &&
286 		    filter->ip_proto == ip_proto &&
287 		    filter->src_port == src_port &&
288 		    filter->dst_port == dst_port) {
289 			ret = filter;
290 			break;
291 		}
292 	}
293 
294 	return ret;
295 }
296 
297 static int
298 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
299 		   u16 rxq_index, u32 flow_id)
300 {
301 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
302 	struct mlx4_en_filter *filter;
303 	const struct iphdr *ip;
304 	const __be16 *ports;
305 	u8 ip_proto;
306 	__be32 src_ip;
307 	__be32 dst_ip;
308 	__be16 src_port;
309 	__be16 dst_port;
310 	int nhoff = skb_network_offset(skb);
311 	int ret = 0;
312 
313 	if (skb->protocol != htons(ETH_P_IP))
314 		return -EPROTONOSUPPORT;
315 
316 	ip = (const struct iphdr *)(skb->data + nhoff);
317 	if (ip_is_fragment(ip))
318 		return -EPROTONOSUPPORT;
319 
320 	if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
321 		return -EPROTONOSUPPORT;
322 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
323 
324 	ip_proto = ip->protocol;
325 	src_ip = ip->saddr;
326 	dst_ip = ip->daddr;
327 	src_port = ports[0];
328 	dst_port = ports[1];
329 
330 	spin_lock_bh(&priv->filters_lock);
331 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
332 				     src_port, dst_port);
333 	if (filter) {
334 		if (filter->rxq_index == rxq_index)
335 			goto out;
336 
337 		filter->rxq_index = rxq_index;
338 	} else {
339 		filter = mlx4_en_filter_alloc(priv, rxq_index,
340 					      src_ip, dst_ip, ip_proto,
341 					      src_port, dst_port, flow_id);
342 		if (!filter) {
343 			ret = -ENOMEM;
344 			goto err;
345 		}
346 	}
347 
348 	queue_work(priv->mdev->workqueue, &filter->work);
349 
350 out:
351 	ret = filter->id;
352 err:
353 	spin_unlock_bh(&priv->filters_lock);
354 
355 	return ret;
356 }
357 
358 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
359 {
360 	struct mlx4_en_filter *filter, *tmp;
361 	LIST_HEAD(del_list);
362 
363 	spin_lock_bh(&priv->filters_lock);
364 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
365 		list_move(&filter->next, &del_list);
366 		hlist_del(&filter->filter_chain);
367 	}
368 	spin_unlock_bh(&priv->filters_lock);
369 
370 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
371 		cancel_work_sync(&filter->work);
372 		mlx4_en_filter_free(filter);
373 	}
374 }
375 
376 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
377 {
378 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
379 	LIST_HEAD(del_list);
380 	int i = 0;
381 
382 	spin_lock_bh(&priv->filters_lock);
383 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
384 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
385 			break;
386 
387 		if (filter->activated &&
388 		    !work_pending(&filter->work) &&
389 		    rps_may_expire_flow(priv->dev,
390 					filter->rxq_index, filter->flow_id,
391 					filter->id)) {
392 			list_move(&filter->next, &del_list);
393 			hlist_del(&filter->filter_chain);
394 		} else
395 			last_filter = filter;
396 
397 		i++;
398 	}
399 
400 	if (last_filter && (&last_filter->next != priv->filters.next))
401 		list_move(&priv->filters, &last_filter->next);
402 
403 	spin_unlock_bh(&priv->filters_lock);
404 
405 	list_for_each_entry_safe(filter, tmp, &del_list, next)
406 		mlx4_en_filter_free(filter);
407 }
408 #endif
409 
410 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
411 				   __be16 proto, u16 vid)
412 {
413 	struct mlx4_en_priv *priv = netdev_priv(dev);
414 	struct mlx4_en_dev *mdev = priv->mdev;
415 	int err;
416 	int idx;
417 
418 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
419 
420 	set_bit(vid, priv->active_vlans);
421 
422 	/* Add VID to port VLAN filter */
423 	mutex_lock(&mdev->state_lock);
424 	if (mdev->device_up && priv->port_up) {
425 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
426 		if (err)
427 			en_err(priv, "Failed configuring VLAN filter\n");
428 	}
429 	if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
430 		en_dbg(HW, priv, "failed adding vlan %d\n", vid);
431 	mutex_unlock(&mdev->state_lock);
432 
433 	return 0;
434 }
435 
436 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
437 				    __be16 proto, u16 vid)
438 {
439 	struct mlx4_en_priv *priv = netdev_priv(dev);
440 	struct mlx4_en_dev *mdev = priv->mdev;
441 	int err;
442 
443 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
444 
445 	clear_bit(vid, priv->active_vlans);
446 
447 	/* Remove VID from port VLAN filter */
448 	mutex_lock(&mdev->state_lock);
449 	mlx4_unregister_vlan(mdev->dev, priv->port, vid);
450 
451 	if (mdev->device_up && priv->port_up) {
452 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
453 		if (err)
454 			en_err(priv, "Failed configuring VLAN filter\n");
455 	}
456 	mutex_unlock(&mdev->state_lock);
457 
458 	return 0;
459 }
460 
461 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
462 {
463 	int i;
464 	for (i = ETH_ALEN - 1; i >= 0; --i) {
465 		dst_mac[i] = src_mac & 0xff;
466 		src_mac >>= 8;
467 	}
468 	memset(&dst_mac[ETH_ALEN], 0, 2);
469 }
470 
471 
472 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
473 				    int qpn, u64 *reg_id)
474 {
475 	int err;
476 	struct mlx4_spec_list spec_eth_outer = { {NULL} };
477 	struct mlx4_spec_list spec_vxlan     = { {NULL} };
478 	struct mlx4_spec_list spec_eth_inner = { {NULL} };
479 
480 	struct mlx4_net_trans_rule rule = {
481 		.queue_mode = MLX4_NET_TRANS_Q_FIFO,
482 		.exclusive = 0,
483 		.allow_loopback = 1,
484 		.promisc_mode = MLX4_FS_REGULAR,
485 		.priority = MLX4_DOMAIN_NIC,
486 	};
487 
488 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
489 
490 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
491 		return 0; /* do nothing */
492 
493 	rule.port = priv->port;
494 	rule.qpn = qpn;
495 	INIT_LIST_HEAD(&rule.list);
496 
497 	spec_eth_outer.id = MLX4_NET_TRANS_RULE_ID_ETH;
498 	memcpy(spec_eth_outer.eth.dst_mac, addr, ETH_ALEN);
499 	memcpy(spec_eth_outer.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
500 
501 	spec_vxlan.id = MLX4_NET_TRANS_RULE_ID_VXLAN;    /* any vxlan header */
502 	spec_eth_inner.id = MLX4_NET_TRANS_RULE_ID_ETH;	 /* any inner eth header */
503 
504 	list_add_tail(&spec_eth_outer.list, &rule.list);
505 	list_add_tail(&spec_vxlan.list,     &rule.list);
506 	list_add_tail(&spec_eth_inner.list, &rule.list);
507 
508 	err = mlx4_flow_attach(priv->mdev->dev, &rule, reg_id);
509 	if (err) {
510 		en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
511 		return err;
512 	}
513 	en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
514 	return 0;
515 }
516 
517 
518 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
519 				unsigned char *mac, int *qpn, u64 *reg_id)
520 {
521 	struct mlx4_en_dev *mdev = priv->mdev;
522 	struct mlx4_dev *dev = mdev->dev;
523 	int err;
524 
525 	switch (dev->caps.steering_mode) {
526 	case MLX4_STEERING_MODE_B0: {
527 		struct mlx4_qp qp;
528 		u8 gid[16] = {0};
529 
530 		qp.qpn = *qpn;
531 		memcpy(&gid[10], mac, ETH_ALEN);
532 		gid[5] = priv->port;
533 
534 		err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
535 		break;
536 	}
537 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
538 		struct mlx4_spec_list spec_eth = { {NULL} };
539 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
540 
541 		struct mlx4_net_trans_rule rule = {
542 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
543 			.exclusive = 0,
544 			.allow_loopback = 1,
545 			.promisc_mode = MLX4_FS_REGULAR,
546 			.priority = MLX4_DOMAIN_NIC,
547 		};
548 
549 		rule.port = priv->port;
550 		rule.qpn = *qpn;
551 		INIT_LIST_HEAD(&rule.list);
552 
553 		spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
554 		memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
555 		memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
556 		list_add_tail(&spec_eth.list, &rule.list);
557 
558 		err = mlx4_flow_attach(dev, &rule, reg_id);
559 		break;
560 	}
561 	default:
562 		return -EINVAL;
563 	}
564 	if (err)
565 		en_warn(priv, "Failed Attaching Unicast\n");
566 
567 	return err;
568 }
569 
570 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
571 				     unsigned char *mac, int qpn, u64 reg_id)
572 {
573 	struct mlx4_en_dev *mdev = priv->mdev;
574 	struct mlx4_dev *dev = mdev->dev;
575 
576 	switch (dev->caps.steering_mode) {
577 	case MLX4_STEERING_MODE_B0: {
578 		struct mlx4_qp qp;
579 		u8 gid[16] = {0};
580 
581 		qp.qpn = qpn;
582 		memcpy(&gid[10], mac, ETH_ALEN);
583 		gid[5] = priv->port;
584 
585 		mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
586 		break;
587 	}
588 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
589 		mlx4_flow_detach(dev, reg_id);
590 		break;
591 	}
592 	default:
593 		en_err(priv, "Invalid steering mode.\n");
594 	}
595 }
596 
597 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
598 {
599 	struct mlx4_en_dev *mdev = priv->mdev;
600 	struct mlx4_dev *dev = mdev->dev;
601 	struct mlx4_mac_entry *entry;
602 	int index = 0;
603 	int err = 0;
604 	u64 reg_id;
605 	int *qpn = &priv->base_qpn;
606 	u64 mac = mlx4_en_mac_to_u64(priv->dev->dev_addr);
607 
608 	en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
609 	       priv->dev->dev_addr);
610 	index = mlx4_register_mac(dev, priv->port, mac);
611 	if (index < 0) {
612 		err = index;
613 		en_err(priv, "Failed adding MAC: %pM\n",
614 		       priv->dev->dev_addr);
615 		return err;
616 	}
617 
618 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
619 		int base_qpn = mlx4_get_base_qpn(dev, priv->port);
620 		*qpn = base_qpn + index;
621 		return 0;
622 	}
623 
624 	err = mlx4_qp_reserve_range(dev, 1, 1, qpn);
625 	en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
626 	if (err) {
627 		en_err(priv, "Failed to reserve qp for mac registration\n");
628 		goto qp_err;
629 	}
630 
631 	err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
632 	if (err)
633 		goto steer_err;
634 
635 	err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
636 				       &priv->tunnel_reg_id);
637 	if (err)
638 		goto tunnel_err;
639 
640 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
641 	if (!entry) {
642 		err = -ENOMEM;
643 		goto alloc_err;
644 	}
645 	memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
646 	entry->reg_id = reg_id;
647 
648 	hlist_add_head_rcu(&entry->hlist,
649 			   &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
650 
651 	return 0;
652 
653 alloc_err:
654 	if (priv->tunnel_reg_id)
655 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
656 tunnel_err:
657 	mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
658 
659 steer_err:
660 	mlx4_qp_release_range(dev, *qpn, 1);
661 
662 qp_err:
663 	mlx4_unregister_mac(dev, priv->port, mac);
664 	return err;
665 }
666 
667 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
668 {
669 	struct mlx4_en_dev *mdev = priv->mdev;
670 	struct mlx4_dev *dev = mdev->dev;
671 	int qpn = priv->base_qpn;
672 	u64 mac;
673 
674 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
675 		mac = mlx4_en_mac_to_u64(priv->dev->dev_addr);
676 		en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
677 		       priv->dev->dev_addr);
678 		mlx4_unregister_mac(dev, priv->port, mac);
679 	} else {
680 		struct mlx4_mac_entry *entry;
681 		struct hlist_node *tmp;
682 		struct hlist_head *bucket;
683 		unsigned int i;
684 
685 		for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
686 			bucket = &priv->mac_hash[i];
687 			hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
688 				mac = mlx4_en_mac_to_u64(entry->mac);
689 				en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
690 				       entry->mac);
691 				mlx4_en_uc_steer_release(priv, entry->mac,
692 							 qpn, entry->reg_id);
693 
694 				mlx4_unregister_mac(dev, priv->port, mac);
695 				hlist_del_rcu(&entry->hlist);
696 				kfree_rcu(entry, rcu);
697 			}
698 		}
699 
700 		if (priv->tunnel_reg_id) {
701 			mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
702 			priv->tunnel_reg_id = 0;
703 		}
704 
705 		en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
706 		       priv->port, qpn);
707 		mlx4_qp_release_range(dev, qpn, 1);
708 		priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
709 	}
710 }
711 
712 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
713 			       unsigned char *new_mac, unsigned char *prev_mac)
714 {
715 	struct mlx4_en_dev *mdev = priv->mdev;
716 	struct mlx4_dev *dev = mdev->dev;
717 	int err = 0;
718 	u64 new_mac_u64 = mlx4_en_mac_to_u64(new_mac);
719 
720 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
721 		struct hlist_head *bucket;
722 		unsigned int mac_hash;
723 		struct mlx4_mac_entry *entry;
724 		struct hlist_node *tmp;
725 		u64 prev_mac_u64 = mlx4_en_mac_to_u64(prev_mac);
726 
727 		bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
728 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
729 			if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
730 				mlx4_en_uc_steer_release(priv, entry->mac,
731 							 qpn, entry->reg_id);
732 				mlx4_unregister_mac(dev, priv->port,
733 						    prev_mac_u64);
734 				hlist_del_rcu(&entry->hlist);
735 				synchronize_rcu();
736 				memcpy(entry->mac, new_mac, ETH_ALEN);
737 				entry->reg_id = 0;
738 				mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
739 				hlist_add_head_rcu(&entry->hlist,
740 						   &priv->mac_hash[mac_hash]);
741 				mlx4_register_mac(dev, priv->port, new_mac_u64);
742 				err = mlx4_en_uc_steer_add(priv, new_mac,
743 							   &qpn,
744 							   &entry->reg_id);
745 				if (err)
746 					return err;
747 				if (priv->tunnel_reg_id) {
748 					mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
749 					priv->tunnel_reg_id = 0;
750 				}
751 				err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
752 							       &priv->tunnel_reg_id);
753 				return err;
754 			}
755 		}
756 		return -EINVAL;
757 	}
758 
759 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
760 }
761 
762 u64 mlx4_en_mac_to_u64(u8 *addr)
763 {
764 	u64 mac = 0;
765 	int i;
766 
767 	for (i = 0; i < ETH_ALEN; i++) {
768 		mac <<= 8;
769 		mac |= addr[i];
770 	}
771 	return mac;
772 }
773 
774 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv)
775 {
776 	int err = 0;
777 
778 	if (priv->port_up) {
779 		/* Remove old MAC and insert the new one */
780 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
781 					  priv->dev->dev_addr, priv->prev_mac);
782 		if (err)
783 			en_err(priv, "Failed changing HW MAC address\n");
784 		memcpy(priv->prev_mac, priv->dev->dev_addr,
785 		       sizeof(priv->prev_mac));
786 	} else
787 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
788 
789 	return err;
790 }
791 
792 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
793 {
794 	struct mlx4_en_priv *priv = netdev_priv(dev);
795 	struct mlx4_en_dev *mdev = priv->mdev;
796 	struct sockaddr *saddr = addr;
797 	int err;
798 
799 	if (!is_valid_ether_addr(saddr->sa_data))
800 		return -EADDRNOTAVAIL;
801 
802 	memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
803 
804 	mutex_lock(&mdev->state_lock);
805 	err = mlx4_en_do_set_mac(priv);
806 	mutex_unlock(&mdev->state_lock);
807 
808 	return err;
809 }
810 
811 static void mlx4_en_clear_list(struct net_device *dev)
812 {
813 	struct mlx4_en_priv *priv = netdev_priv(dev);
814 	struct mlx4_en_mc_list *tmp, *mc_to_del;
815 
816 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
817 		list_del(&mc_to_del->list);
818 		kfree(mc_to_del);
819 	}
820 }
821 
822 static void mlx4_en_cache_mclist(struct net_device *dev)
823 {
824 	struct mlx4_en_priv *priv = netdev_priv(dev);
825 	struct netdev_hw_addr *ha;
826 	struct mlx4_en_mc_list *tmp;
827 
828 	mlx4_en_clear_list(dev);
829 	netdev_for_each_mc_addr(ha, dev) {
830 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
831 		if (!tmp) {
832 			mlx4_en_clear_list(dev);
833 			return;
834 		}
835 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
836 		list_add_tail(&tmp->list, &priv->mc_list);
837 	}
838 }
839 
840 static void update_mclist_flags(struct mlx4_en_priv *priv,
841 				struct list_head *dst,
842 				struct list_head *src)
843 {
844 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
845 	bool found;
846 
847 	/* Find all the entries that should be removed from dst,
848 	 * These are the entries that are not found in src
849 	 */
850 	list_for_each_entry(dst_tmp, dst, list) {
851 		found = false;
852 		list_for_each_entry(src_tmp, src, list) {
853 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
854 				found = true;
855 				break;
856 			}
857 		}
858 		if (!found)
859 			dst_tmp->action = MCLIST_REM;
860 	}
861 
862 	/* Add entries that exist in src but not in dst
863 	 * mark them as need to add
864 	 */
865 	list_for_each_entry(src_tmp, src, list) {
866 		found = false;
867 		list_for_each_entry(dst_tmp, dst, list) {
868 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
869 				dst_tmp->action = MCLIST_NONE;
870 				found = true;
871 				break;
872 			}
873 		}
874 		if (!found) {
875 			new_mc = kmemdup(src_tmp,
876 					 sizeof(struct mlx4_en_mc_list),
877 					 GFP_KERNEL);
878 			if (!new_mc)
879 				return;
880 
881 			new_mc->action = MCLIST_ADD;
882 			list_add_tail(&new_mc->list, dst);
883 		}
884 	}
885 }
886 
887 static void mlx4_en_set_rx_mode(struct net_device *dev)
888 {
889 	struct mlx4_en_priv *priv = netdev_priv(dev);
890 
891 	if (!priv->port_up)
892 		return;
893 
894 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
895 }
896 
897 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
898 				     struct mlx4_en_dev *mdev)
899 {
900 	int err = 0;
901 
902 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
903 		if (netif_msg_rx_status(priv))
904 			en_warn(priv, "Entering promiscuous mode\n");
905 		priv->flags |= MLX4_EN_FLAG_PROMISC;
906 
907 		/* Enable promiscouos mode */
908 		switch (mdev->dev->caps.steering_mode) {
909 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
910 			err = mlx4_flow_steer_promisc_add(mdev->dev,
911 							  priv->port,
912 							  priv->base_qpn,
913 							  MLX4_FS_ALL_DEFAULT);
914 			if (err)
915 				en_err(priv, "Failed enabling promiscuous mode\n");
916 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
917 			break;
918 
919 		case MLX4_STEERING_MODE_B0:
920 			err = mlx4_unicast_promisc_add(mdev->dev,
921 						       priv->base_qpn,
922 						       priv->port);
923 			if (err)
924 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
925 
926 			/* Add the default qp number as multicast
927 			 * promisc
928 			 */
929 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
930 				err = mlx4_multicast_promisc_add(mdev->dev,
931 								 priv->base_qpn,
932 								 priv->port);
933 				if (err)
934 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
935 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
936 			}
937 			break;
938 
939 		case MLX4_STEERING_MODE_A0:
940 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
941 						     priv->port,
942 						     priv->base_qpn,
943 						     1);
944 			if (err)
945 				en_err(priv, "Failed enabling promiscuous mode\n");
946 			break;
947 		}
948 
949 		/* Disable port multicast filter (unconditionally) */
950 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
951 					  0, MLX4_MCAST_DISABLE);
952 		if (err)
953 			en_err(priv, "Failed disabling multicast filter\n");
954 
955 		/* Disable port VLAN filter */
956 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
957 		if (err)
958 			en_err(priv, "Failed disabling VLAN filter\n");
959 	}
960 }
961 
962 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
963 				       struct mlx4_en_dev *mdev)
964 {
965 	int err = 0;
966 
967 	if (netif_msg_rx_status(priv))
968 		en_warn(priv, "Leaving promiscuous mode\n");
969 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
970 
971 	/* Disable promiscouos mode */
972 	switch (mdev->dev->caps.steering_mode) {
973 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
974 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
975 						     priv->port,
976 						     MLX4_FS_ALL_DEFAULT);
977 		if (err)
978 			en_err(priv, "Failed disabling promiscuous mode\n");
979 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
980 		break;
981 
982 	case MLX4_STEERING_MODE_B0:
983 		err = mlx4_unicast_promisc_remove(mdev->dev,
984 						  priv->base_qpn,
985 						  priv->port);
986 		if (err)
987 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
988 		/* Disable Multicast promisc */
989 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
990 			err = mlx4_multicast_promisc_remove(mdev->dev,
991 							    priv->base_qpn,
992 							    priv->port);
993 			if (err)
994 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
995 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
996 		}
997 		break;
998 
999 	case MLX4_STEERING_MODE_A0:
1000 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
1001 					     priv->port,
1002 					     priv->base_qpn, 0);
1003 		if (err)
1004 			en_err(priv, "Failed disabling promiscuous mode\n");
1005 		break;
1006 	}
1007 
1008 	/* Enable port VLAN filter */
1009 	err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
1010 	if (err)
1011 		en_err(priv, "Failed enabling VLAN filter\n");
1012 }
1013 
1014 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
1015 				 struct net_device *dev,
1016 				 struct mlx4_en_dev *mdev)
1017 {
1018 	struct mlx4_en_mc_list *mclist, *tmp;
1019 	u64 mcast_addr = 0;
1020 	u8 mc_list[16] = {0};
1021 	int err = 0;
1022 
1023 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
1024 	if (dev->flags & IFF_ALLMULTI) {
1025 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1026 					  0, MLX4_MCAST_DISABLE);
1027 		if (err)
1028 			en_err(priv, "Failed disabling multicast filter\n");
1029 
1030 		/* Add the default qp number as multicast promisc */
1031 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1032 			switch (mdev->dev->caps.steering_mode) {
1033 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1034 				err = mlx4_flow_steer_promisc_add(mdev->dev,
1035 								  priv->port,
1036 								  priv->base_qpn,
1037 								  MLX4_FS_MC_DEFAULT);
1038 				break;
1039 
1040 			case MLX4_STEERING_MODE_B0:
1041 				err = mlx4_multicast_promisc_add(mdev->dev,
1042 								 priv->base_qpn,
1043 								 priv->port);
1044 				break;
1045 
1046 			case MLX4_STEERING_MODE_A0:
1047 				break;
1048 			}
1049 			if (err)
1050 				en_err(priv, "Failed entering multicast promisc mode\n");
1051 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1052 		}
1053 	} else {
1054 		/* Disable Multicast promisc */
1055 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1056 			switch (mdev->dev->caps.steering_mode) {
1057 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1058 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
1059 								     priv->port,
1060 								     MLX4_FS_MC_DEFAULT);
1061 				break;
1062 
1063 			case MLX4_STEERING_MODE_B0:
1064 				err = mlx4_multicast_promisc_remove(mdev->dev,
1065 								    priv->base_qpn,
1066 								    priv->port);
1067 				break;
1068 
1069 			case MLX4_STEERING_MODE_A0:
1070 				break;
1071 			}
1072 			if (err)
1073 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
1074 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1075 		}
1076 
1077 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1078 					  0, MLX4_MCAST_DISABLE);
1079 		if (err)
1080 			en_err(priv, "Failed disabling multicast filter\n");
1081 
1082 		/* Flush mcast filter and init it with broadcast address */
1083 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1084 				    1, MLX4_MCAST_CONFIG);
1085 
1086 		/* Update multicast list - we cache all addresses so they won't
1087 		 * change while HW is updated holding the command semaphor */
1088 		netif_addr_lock_bh(dev);
1089 		mlx4_en_cache_mclist(dev);
1090 		netif_addr_unlock_bh(dev);
1091 		list_for_each_entry(mclist, &priv->mc_list, list) {
1092 			mcast_addr = mlx4_en_mac_to_u64(mclist->addr);
1093 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1094 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
1095 		}
1096 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1097 					  0, MLX4_MCAST_ENABLE);
1098 		if (err)
1099 			en_err(priv, "Failed enabling multicast filter\n");
1100 
1101 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1102 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1103 			if (mclist->action == MCLIST_REM) {
1104 				/* detach this address and delete from list */
1105 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1106 				mc_list[5] = priv->port;
1107 				err = mlx4_multicast_detach(mdev->dev,
1108 							    &priv->rss_map.indir_qp,
1109 							    mc_list,
1110 							    MLX4_PROT_ETH,
1111 							    mclist->reg_id);
1112 				if (err)
1113 					en_err(priv, "Fail to detach multicast address\n");
1114 
1115 				if (mclist->tunnel_reg_id) {
1116 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1117 					if (err)
1118 						en_err(priv, "Failed to detach multicast address\n");
1119 				}
1120 
1121 				/* remove from list */
1122 				list_del(&mclist->list);
1123 				kfree(mclist);
1124 			} else if (mclist->action == MCLIST_ADD) {
1125 				/* attach the address */
1126 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1127 				/* needed for B0 steering support */
1128 				mc_list[5] = priv->port;
1129 				err = mlx4_multicast_attach(mdev->dev,
1130 							    &priv->rss_map.indir_qp,
1131 							    mc_list,
1132 							    priv->port, 0,
1133 							    MLX4_PROT_ETH,
1134 							    &mclist->reg_id);
1135 				if (err)
1136 					en_err(priv, "Fail to attach multicast address\n");
1137 
1138 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1139 							       &mclist->tunnel_reg_id);
1140 				if (err)
1141 					en_err(priv, "Failed to attach multicast address\n");
1142 			}
1143 		}
1144 	}
1145 }
1146 
1147 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1148 				 struct net_device *dev,
1149 				 struct mlx4_en_dev *mdev)
1150 {
1151 	struct netdev_hw_addr *ha;
1152 	struct mlx4_mac_entry *entry;
1153 	struct hlist_node *tmp;
1154 	bool found;
1155 	u64 mac;
1156 	int err = 0;
1157 	struct hlist_head *bucket;
1158 	unsigned int i;
1159 	int removed = 0;
1160 	u32 prev_flags;
1161 
1162 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1163 	 * since all modification code is protected by mdev->state_lock
1164 	 */
1165 
1166 	/* find what to remove */
1167 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1168 		bucket = &priv->mac_hash[i];
1169 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1170 			found = false;
1171 			netdev_for_each_uc_addr(ha, dev) {
1172 				if (ether_addr_equal_64bits(entry->mac,
1173 							    ha->addr)) {
1174 					found = true;
1175 					break;
1176 				}
1177 			}
1178 
1179 			/* MAC address of the port is not in uc list */
1180 			if (ether_addr_equal_64bits(entry->mac, dev->dev_addr))
1181 				found = true;
1182 
1183 			if (!found) {
1184 				mac = mlx4_en_mac_to_u64(entry->mac);
1185 				mlx4_en_uc_steer_release(priv, entry->mac,
1186 							 priv->base_qpn,
1187 							 entry->reg_id);
1188 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1189 
1190 				hlist_del_rcu(&entry->hlist);
1191 				kfree_rcu(entry, rcu);
1192 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1193 				       entry->mac, priv->port);
1194 				++removed;
1195 			}
1196 		}
1197 	}
1198 
1199 	/* if we didn't remove anything, there is no use in trying to add
1200 	 * again once we are in a forced promisc mode state
1201 	 */
1202 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1203 		return;
1204 
1205 	prev_flags = priv->flags;
1206 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1207 
1208 	/* find what to add */
1209 	netdev_for_each_uc_addr(ha, dev) {
1210 		found = false;
1211 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1212 		hlist_for_each_entry(entry, bucket, hlist) {
1213 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1214 				found = true;
1215 				break;
1216 			}
1217 		}
1218 
1219 		if (!found) {
1220 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1221 			if (!entry) {
1222 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1223 				       ha->addr, priv->port);
1224 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1225 				break;
1226 			}
1227 			mac = mlx4_en_mac_to_u64(ha->addr);
1228 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1229 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1230 			if (err < 0) {
1231 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1232 				       ha->addr, priv->port, err);
1233 				kfree(entry);
1234 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1235 				break;
1236 			}
1237 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1238 						   &priv->base_qpn,
1239 						   &entry->reg_id);
1240 			if (err) {
1241 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1242 				       ha->addr, priv->port, err);
1243 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1244 				kfree(entry);
1245 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1246 				break;
1247 			} else {
1248 				unsigned int mac_hash;
1249 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1250 				       ha->addr, priv->port);
1251 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1252 				bucket = &priv->mac_hash[mac_hash];
1253 				hlist_add_head_rcu(&entry->hlist, bucket);
1254 			}
1255 		}
1256 	}
1257 
1258 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1259 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1260 			priv->port);
1261 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1262 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1263 			priv->port);
1264 	}
1265 }
1266 
1267 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1268 {
1269 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1270 						 rx_mode_task);
1271 	struct mlx4_en_dev *mdev = priv->mdev;
1272 	struct net_device *dev = priv->dev;
1273 
1274 	mutex_lock(&mdev->state_lock);
1275 	if (!mdev->device_up) {
1276 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1277 		goto out;
1278 	}
1279 	if (!priv->port_up) {
1280 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1281 		goto out;
1282 	}
1283 
1284 	if (!netif_carrier_ok(dev)) {
1285 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1286 			if (priv->port_state.link_state) {
1287 				priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1288 				netif_carrier_on(dev);
1289 				en_dbg(LINK, priv, "Link Up\n");
1290 			}
1291 		}
1292 	}
1293 
1294 	if (dev->priv_flags & IFF_UNICAST_FLT)
1295 		mlx4_en_do_uc_filter(priv, dev, mdev);
1296 
1297 	/* Promsicuous mode: disable all filters */
1298 	if ((dev->flags & IFF_PROMISC) ||
1299 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1300 		mlx4_en_set_promisc_mode(priv, mdev);
1301 		goto out;
1302 	}
1303 
1304 	/* Not in promiscuous mode */
1305 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1306 		mlx4_en_clear_promisc_mode(priv, mdev);
1307 
1308 	mlx4_en_do_multicast(priv, dev, mdev);
1309 out:
1310 	mutex_unlock(&mdev->state_lock);
1311 }
1312 
1313 #ifdef CONFIG_NET_POLL_CONTROLLER
1314 static void mlx4_en_netpoll(struct net_device *dev)
1315 {
1316 	struct mlx4_en_priv *priv = netdev_priv(dev);
1317 	struct mlx4_en_cq *cq;
1318 	unsigned long flags;
1319 	int i;
1320 
1321 	for (i = 0; i < priv->rx_ring_num; i++) {
1322 		cq = priv->rx_cq[i];
1323 		spin_lock_irqsave(&cq->lock, flags);
1324 		napi_synchronize(&cq->napi);
1325 		mlx4_en_process_rx_cq(dev, cq, 0);
1326 		spin_unlock_irqrestore(&cq->lock, flags);
1327 	}
1328 }
1329 #endif
1330 
1331 static void mlx4_en_tx_timeout(struct net_device *dev)
1332 {
1333 	struct mlx4_en_priv *priv = netdev_priv(dev);
1334 	struct mlx4_en_dev *mdev = priv->mdev;
1335 	int i;
1336 
1337 	if (netif_msg_timer(priv))
1338 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1339 
1340 	for (i = 0; i < priv->tx_ring_num; i++) {
1341 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1342 			continue;
1343 		en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1344 			i, priv->tx_ring[i]->qpn, priv->tx_ring[i]->cqn,
1345 			priv->tx_ring[i]->cons, priv->tx_ring[i]->prod);
1346 	}
1347 
1348 	priv->port_stats.tx_timeout++;
1349 	en_dbg(DRV, priv, "Scheduling watchdog\n");
1350 	queue_work(mdev->workqueue, &priv->watchdog_task);
1351 }
1352 
1353 
1354 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
1355 {
1356 	struct mlx4_en_priv *priv = netdev_priv(dev);
1357 
1358 	spin_lock_bh(&priv->stats_lock);
1359 	memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
1360 	spin_unlock_bh(&priv->stats_lock);
1361 
1362 	return &priv->ret_stats;
1363 }
1364 
1365 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1366 {
1367 	struct mlx4_en_cq *cq;
1368 	int i;
1369 
1370 	/* If we haven't received a specific coalescing setting
1371 	 * (module param), we set the moderation parameters as follows:
1372 	 * - moder_cnt is set to the number of mtu sized packets to
1373 	 *   satisfy our coalescing target.
1374 	 * - moder_time is set to a fixed value.
1375 	 */
1376 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1377 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1378 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1379 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1380 	en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1381 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1382 
1383 	/* Setup cq moderation params */
1384 	for (i = 0; i < priv->rx_ring_num; i++) {
1385 		cq = priv->rx_cq[i];
1386 		cq->moder_cnt = priv->rx_frames;
1387 		cq->moder_time = priv->rx_usecs;
1388 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1389 		priv->last_moder_packets[i] = 0;
1390 		priv->last_moder_bytes[i] = 0;
1391 	}
1392 
1393 	for (i = 0; i < priv->tx_ring_num; i++) {
1394 		cq = priv->tx_cq[i];
1395 		cq->moder_cnt = priv->tx_frames;
1396 		cq->moder_time = priv->tx_usecs;
1397 	}
1398 
1399 	/* Reset auto-moderation params */
1400 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1401 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1402 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1403 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1404 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1405 	priv->adaptive_rx_coal = 1;
1406 	priv->last_moder_jiffies = 0;
1407 	priv->last_moder_tx_packets = 0;
1408 }
1409 
1410 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1411 {
1412 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1413 	struct mlx4_en_cq *cq;
1414 	unsigned long packets;
1415 	unsigned long rate;
1416 	unsigned long avg_pkt_size;
1417 	unsigned long rx_packets;
1418 	unsigned long rx_bytes;
1419 	unsigned long rx_pkt_diff;
1420 	int moder_time;
1421 	int ring, err;
1422 
1423 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1424 		return;
1425 
1426 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1427 		spin_lock_bh(&priv->stats_lock);
1428 		rx_packets = priv->rx_ring[ring]->packets;
1429 		rx_bytes = priv->rx_ring[ring]->bytes;
1430 		spin_unlock_bh(&priv->stats_lock);
1431 
1432 		rx_pkt_diff = ((unsigned long) (rx_packets -
1433 				priv->last_moder_packets[ring]));
1434 		packets = rx_pkt_diff;
1435 		rate = packets * HZ / period;
1436 		avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1437 				priv->last_moder_bytes[ring])) / packets : 0;
1438 
1439 		/* Apply auto-moderation only when packet rate
1440 		 * exceeds a rate that it matters */
1441 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1442 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1443 			if (rate < priv->pkt_rate_low)
1444 				moder_time = priv->rx_usecs_low;
1445 			else if (rate > priv->pkt_rate_high)
1446 				moder_time = priv->rx_usecs_high;
1447 			else
1448 				moder_time = (rate - priv->pkt_rate_low) *
1449 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1450 					(priv->pkt_rate_high - priv->pkt_rate_low) +
1451 					priv->rx_usecs_low;
1452 		} else {
1453 			moder_time = priv->rx_usecs_low;
1454 		}
1455 
1456 		if (moder_time != priv->last_moder_time[ring]) {
1457 			priv->last_moder_time[ring] = moder_time;
1458 			cq = priv->rx_cq[ring];
1459 			cq->moder_time = moder_time;
1460 			cq->moder_cnt = priv->rx_frames;
1461 			err = mlx4_en_set_cq_moder(priv, cq);
1462 			if (err)
1463 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1464 				       ring);
1465 		}
1466 		priv->last_moder_packets[ring] = rx_packets;
1467 		priv->last_moder_bytes[ring] = rx_bytes;
1468 	}
1469 
1470 	priv->last_moder_jiffies = jiffies;
1471 }
1472 
1473 static void mlx4_en_do_get_stats(struct work_struct *work)
1474 {
1475 	struct delayed_work *delay = to_delayed_work(work);
1476 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1477 						 stats_task);
1478 	struct mlx4_en_dev *mdev = priv->mdev;
1479 	int err;
1480 
1481 	mutex_lock(&mdev->state_lock);
1482 	if (mdev->device_up) {
1483 		if (priv->port_up) {
1484 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1485 			if (err)
1486 				en_dbg(HW, priv, "Could not update stats\n");
1487 
1488 			mlx4_en_auto_moderation(priv);
1489 		}
1490 
1491 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1492 	}
1493 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1494 		mlx4_en_do_set_mac(priv);
1495 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1496 	}
1497 	mutex_unlock(&mdev->state_lock);
1498 }
1499 
1500 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1501  * periodically
1502  */
1503 static void mlx4_en_service_task(struct work_struct *work)
1504 {
1505 	struct delayed_work *delay = to_delayed_work(work);
1506 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1507 						 service_task);
1508 	struct mlx4_en_dev *mdev = priv->mdev;
1509 
1510 	mutex_lock(&mdev->state_lock);
1511 	if (mdev->device_up) {
1512 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1513 			mlx4_en_ptp_overflow_check(mdev);
1514 
1515 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1516 				   SERVICE_TASK_DELAY);
1517 	}
1518 	mutex_unlock(&mdev->state_lock);
1519 }
1520 
1521 static void mlx4_en_linkstate(struct work_struct *work)
1522 {
1523 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1524 						 linkstate_task);
1525 	struct mlx4_en_dev *mdev = priv->mdev;
1526 	int linkstate = priv->link_state;
1527 
1528 	mutex_lock(&mdev->state_lock);
1529 	/* If observable port state changed set carrier state and
1530 	 * report to system log */
1531 	if (priv->last_link_state != linkstate) {
1532 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1533 			en_info(priv, "Link Down\n");
1534 			netif_carrier_off(priv->dev);
1535 		} else {
1536 			en_info(priv, "Link Up\n");
1537 			netif_carrier_on(priv->dev);
1538 		}
1539 	}
1540 	priv->last_link_state = linkstate;
1541 	mutex_unlock(&mdev->state_lock);
1542 }
1543 
1544 
1545 int mlx4_en_start_port(struct net_device *dev)
1546 {
1547 	struct mlx4_en_priv *priv = netdev_priv(dev);
1548 	struct mlx4_en_dev *mdev = priv->mdev;
1549 	struct mlx4_en_cq *cq;
1550 	struct mlx4_en_tx_ring *tx_ring;
1551 	int rx_index = 0;
1552 	int tx_index = 0;
1553 	int err = 0;
1554 	int i;
1555 	int j;
1556 	u8 mc_list[16] = {0};
1557 
1558 	if (priv->port_up) {
1559 		en_dbg(DRV, priv, "start port called while port already up\n");
1560 		return 0;
1561 	}
1562 
1563 	INIT_LIST_HEAD(&priv->mc_list);
1564 	INIT_LIST_HEAD(&priv->curr_list);
1565 	INIT_LIST_HEAD(&priv->ethtool_list);
1566 	memset(&priv->ethtool_rules[0], 0,
1567 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1568 
1569 	/* Calculate Rx buf size */
1570 	dev->mtu = min(dev->mtu, priv->max_mtu);
1571 	mlx4_en_calc_rx_buf(dev);
1572 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1573 
1574 	/* Configure rx cq's and rings */
1575 	err = mlx4_en_activate_rx_rings(priv);
1576 	if (err) {
1577 		en_err(priv, "Failed to activate RX rings\n");
1578 		return err;
1579 	}
1580 	for (i = 0; i < priv->rx_ring_num; i++) {
1581 		cq = priv->rx_cq[i];
1582 
1583 		mlx4_en_cq_init_lock(cq);
1584 
1585 		err = mlx4_en_activate_cq(priv, cq, i);
1586 		if (err) {
1587 			en_err(priv, "Failed activating Rx CQ\n");
1588 			goto cq_err;
1589 		}
1590 		for (j = 0; j < cq->size; j++)
1591 			cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1592 		err = mlx4_en_set_cq_moder(priv, cq);
1593 		if (err) {
1594 			en_err(priv, "Failed setting cq moderation parameters");
1595 			mlx4_en_deactivate_cq(priv, cq);
1596 			goto cq_err;
1597 		}
1598 		mlx4_en_arm_cq(priv, cq);
1599 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1600 		++rx_index;
1601 	}
1602 
1603 	/* Set qp number */
1604 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1605 	err = mlx4_en_get_qp(priv);
1606 	if (err) {
1607 		en_err(priv, "Failed getting eth qp\n");
1608 		goto cq_err;
1609 	}
1610 	mdev->mac_removed[priv->port] = 0;
1611 
1612 	err = mlx4_en_config_rss_steer(priv);
1613 	if (err) {
1614 		en_err(priv, "Failed configuring rss steering\n");
1615 		goto mac_err;
1616 	}
1617 
1618 	err = mlx4_en_create_drop_qp(priv);
1619 	if (err)
1620 		goto rss_err;
1621 
1622 	/* Configure tx cq's and rings */
1623 	for (i = 0; i < priv->tx_ring_num; i++) {
1624 		/* Configure cq */
1625 		cq = priv->tx_cq[i];
1626 		err = mlx4_en_activate_cq(priv, cq, i);
1627 		if (err) {
1628 			en_err(priv, "Failed allocating Tx CQ\n");
1629 			goto tx_err;
1630 		}
1631 		err = mlx4_en_set_cq_moder(priv, cq);
1632 		if (err) {
1633 			en_err(priv, "Failed setting cq moderation parameters");
1634 			mlx4_en_deactivate_cq(priv, cq);
1635 			goto tx_err;
1636 		}
1637 		en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1638 		cq->buf->wqe_index = cpu_to_be16(0xffff);
1639 
1640 		/* Configure ring */
1641 		tx_ring = priv->tx_ring[i];
1642 		err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1643 			i / priv->num_tx_rings_p_up);
1644 		if (err) {
1645 			en_err(priv, "Failed allocating Tx ring\n");
1646 			mlx4_en_deactivate_cq(priv, cq);
1647 			goto tx_err;
1648 		}
1649 		tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1650 
1651 		/* Arm CQ for TX completions */
1652 		mlx4_en_arm_cq(priv, cq);
1653 
1654 		/* Set initial ownership of all Tx TXBBs to SW (1) */
1655 		for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1656 			*((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1657 		++tx_index;
1658 	}
1659 
1660 	/* Configure port */
1661 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1662 				    priv->rx_skb_size + ETH_FCS_LEN,
1663 				    priv->prof->tx_pause,
1664 				    priv->prof->tx_ppp,
1665 				    priv->prof->rx_pause,
1666 				    priv->prof->rx_ppp);
1667 	if (err) {
1668 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1669 		       priv->port, err);
1670 		goto tx_err;
1671 	}
1672 	/* Set default qp number */
1673 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1674 	if (err) {
1675 		en_err(priv, "Failed setting default qp numbers\n");
1676 		goto tx_err;
1677 	}
1678 
1679 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1680 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC);
1681 		if (err) {
1682 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1683 			       err);
1684 			goto tx_err;
1685 		}
1686 	}
1687 
1688 	/* Init port */
1689 	en_dbg(HW, priv, "Initializing port\n");
1690 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1691 	if (err) {
1692 		en_err(priv, "Failed Initializing port\n");
1693 		goto tx_err;
1694 	}
1695 
1696 	/* Attach rx QP to bradcast address */
1697 	memset(&mc_list[10], 0xff, ETH_ALEN);
1698 	mc_list[5] = priv->port; /* needed for B0 steering support */
1699 	if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1700 				  priv->port, 0, MLX4_PROT_ETH,
1701 				  &priv->broadcast_id))
1702 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1703 
1704 	/* Must redo promiscuous mode setup. */
1705 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1706 
1707 	/* Schedule multicast task to populate multicast list */
1708 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1709 
1710 	mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
1711 
1712 	priv->port_up = true;
1713 	netif_tx_start_all_queues(dev);
1714 	netif_device_attach(dev);
1715 
1716 	return 0;
1717 
1718 tx_err:
1719 	while (tx_index--) {
1720 		mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1721 		mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1722 	}
1723 	mlx4_en_destroy_drop_qp(priv);
1724 rss_err:
1725 	mlx4_en_release_rss_steer(priv);
1726 mac_err:
1727 	mlx4_en_put_qp(priv);
1728 cq_err:
1729 	while (rx_index--)
1730 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1731 	for (i = 0; i < priv->rx_ring_num; i++)
1732 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1733 
1734 	return err; /* need to close devices */
1735 }
1736 
1737 
1738 void mlx4_en_stop_port(struct net_device *dev, int detach)
1739 {
1740 	struct mlx4_en_priv *priv = netdev_priv(dev);
1741 	struct mlx4_en_dev *mdev = priv->mdev;
1742 	struct mlx4_en_mc_list *mclist, *tmp;
1743 	struct ethtool_flow_id *flow, *tmp_flow;
1744 	int i;
1745 	u8 mc_list[16] = {0};
1746 
1747 	if (!priv->port_up) {
1748 		en_dbg(DRV, priv, "stop port called while port already down\n");
1749 		return;
1750 	}
1751 
1752 	/* close port*/
1753 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1754 
1755 	/* Synchronize with tx routine */
1756 	netif_tx_lock_bh(dev);
1757 	if (detach)
1758 		netif_device_detach(dev);
1759 	netif_tx_stop_all_queues(dev);
1760 	netif_tx_unlock_bh(dev);
1761 
1762 	netif_tx_disable(dev);
1763 
1764 	/* Set port as not active */
1765 	priv->port_up = false;
1766 
1767 	/* Promsicuous mode */
1768 	if (mdev->dev->caps.steering_mode ==
1769 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1770 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1771 				 MLX4_EN_FLAG_MC_PROMISC);
1772 		mlx4_flow_steer_promisc_remove(mdev->dev,
1773 					       priv->port,
1774 					       MLX4_FS_ALL_DEFAULT);
1775 		mlx4_flow_steer_promisc_remove(mdev->dev,
1776 					       priv->port,
1777 					       MLX4_FS_MC_DEFAULT);
1778 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1779 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1780 
1781 		/* Disable promiscouos mode */
1782 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1783 					    priv->port);
1784 
1785 		/* Disable Multicast promisc */
1786 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1787 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1788 						      priv->port);
1789 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1790 		}
1791 	}
1792 
1793 	/* Detach All multicasts */
1794 	memset(&mc_list[10], 0xff, ETH_ALEN);
1795 	mc_list[5] = priv->port; /* needed for B0 steering support */
1796 	mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1797 			      MLX4_PROT_ETH, priv->broadcast_id);
1798 	list_for_each_entry(mclist, &priv->curr_list, list) {
1799 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1800 		mc_list[5] = priv->port;
1801 		mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1802 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1803 		if (mclist->tunnel_reg_id)
1804 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1805 	}
1806 	mlx4_en_clear_list(dev);
1807 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1808 		list_del(&mclist->list);
1809 		kfree(mclist);
1810 	}
1811 
1812 	/* Flush multicast filter */
1813 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1814 
1815 	/* Remove flow steering rules for the port*/
1816 	if (mdev->dev->caps.steering_mode ==
1817 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1818 		ASSERT_RTNL();
1819 		list_for_each_entry_safe(flow, tmp_flow,
1820 					 &priv->ethtool_list, list) {
1821 			mlx4_flow_detach(mdev->dev, flow->id);
1822 			list_del(&flow->list);
1823 		}
1824 	}
1825 
1826 	mlx4_en_destroy_drop_qp(priv);
1827 
1828 	/* Free TX Rings */
1829 	for (i = 0; i < priv->tx_ring_num; i++) {
1830 		mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1831 		mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1832 	}
1833 	msleep(10);
1834 
1835 	for (i = 0; i < priv->tx_ring_num; i++)
1836 		mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1837 
1838 	/* Free RSS qps */
1839 	mlx4_en_release_rss_steer(priv);
1840 
1841 	/* Unregister Mac address for the port */
1842 	mlx4_en_put_qp(priv);
1843 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1844 		mdev->mac_removed[priv->port] = 1;
1845 
1846 	/* Free RX Rings */
1847 	for (i = 0; i < priv->rx_ring_num; i++) {
1848 		struct mlx4_en_cq *cq = priv->rx_cq[i];
1849 
1850 		local_bh_disable();
1851 		while (!mlx4_en_cq_lock_napi(cq)) {
1852 			pr_info("CQ %d locked\n", i);
1853 			mdelay(1);
1854 		}
1855 		local_bh_enable();
1856 
1857 		while (test_bit(NAPI_STATE_SCHED, &cq->napi.state))
1858 			msleep(1);
1859 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1860 		mlx4_en_deactivate_cq(priv, cq);
1861 	}
1862 }
1863 
1864 static void mlx4_en_restart(struct work_struct *work)
1865 {
1866 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1867 						 watchdog_task);
1868 	struct mlx4_en_dev *mdev = priv->mdev;
1869 	struct net_device *dev = priv->dev;
1870 
1871 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1872 
1873 	mutex_lock(&mdev->state_lock);
1874 	if (priv->port_up) {
1875 		mlx4_en_stop_port(dev, 1);
1876 		if (mlx4_en_start_port(dev))
1877 			en_err(priv, "Failed restarting port %d\n", priv->port);
1878 	}
1879 	mutex_unlock(&mdev->state_lock);
1880 }
1881 
1882 static void mlx4_en_clear_stats(struct net_device *dev)
1883 {
1884 	struct mlx4_en_priv *priv = netdev_priv(dev);
1885 	struct mlx4_en_dev *mdev = priv->mdev;
1886 	int i;
1887 
1888 	if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1889 		en_dbg(HW, priv, "Failed dumping statistics\n");
1890 
1891 	memset(&priv->stats, 0, sizeof(priv->stats));
1892 	memset(&priv->pstats, 0, sizeof(priv->pstats));
1893 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1894 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1895 
1896 	for (i = 0; i < priv->tx_ring_num; i++) {
1897 		priv->tx_ring[i]->bytes = 0;
1898 		priv->tx_ring[i]->packets = 0;
1899 		priv->tx_ring[i]->tx_csum = 0;
1900 	}
1901 	for (i = 0; i < priv->rx_ring_num; i++) {
1902 		priv->rx_ring[i]->bytes = 0;
1903 		priv->rx_ring[i]->packets = 0;
1904 		priv->rx_ring[i]->csum_ok = 0;
1905 		priv->rx_ring[i]->csum_none = 0;
1906 	}
1907 }
1908 
1909 static int mlx4_en_open(struct net_device *dev)
1910 {
1911 	struct mlx4_en_priv *priv = netdev_priv(dev);
1912 	struct mlx4_en_dev *mdev = priv->mdev;
1913 	int err = 0;
1914 
1915 	mutex_lock(&mdev->state_lock);
1916 
1917 	if (!mdev->device_up) {
1918 		en_err(priv, "Cannot open - device down/disabled\n");
1919 		err = -EBUSY;
1920 		goto out;
1921 	}
1922 
1923 	/* Reset HW statistics and SW counters */
1924 	mlx4_en_clear_stats(dev);
1925 
1926 	err = mlx4_en_start_port(dev);
1927 	if (err)
1928 		en_err(priv, "Failed starting port:%d\n", priv->port);
1929 
1930 out:
1931 	mutex_unlock(&mdev->state_lock);
1932 	return err;
1933 }
1934 
1935 
1936 static int mlx4_en_close(struct net_device *dev)
1937 {
1938 	struct mlx4_en_priv *priv = netdev_priv(dev);
1939 	struct mlx4_en_dev *mdev = priv->mdev;
1940 
1941 	en_dbg(IFDOWN, priv, "Close port called\n");
1942 
1943 	mutex_lock(&mdev->state_lock);
1944 
1945 	mlx4_en_stop_port(dev, 0);
1946 	netif_carrier_off(dev);
1947 
1948 	mutex_unlock(&mdev->state_lock);
1949 	return 0;
1950 }
1951 
1952 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1953 {
1954 	int i;
1955 
1956 #ifdef CONFIG_RFS_ACCEL
1957 	free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1958 	priv->dev->rx_cpu_rmap = NULL;
1959 #endif
1960 
1961 	for (i = 0; i < priv->tx_ring_num; i++) {
1962 		if (priv->tx_ring && priv->tx_ring[i])
1963 			mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1964 		if (priv->tx_cq && priv->tx_cq[i])
1965 			mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1966 	}
1967 
1968 	for (i = 0; i < priv->rx_ring_num; i++) {
1969 		if (priv->rx_ring[i])
1970 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1971 				priv->prof->rx_ring_size, priv->stride);
1972 		if (priv->rx_cq[i])
1973 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1974 	}
1975 
1976 	if (priv->base_tx_qpn) {
1977 		mlx4_qp_release_range(priv->mdev->dev, priv->base_tx_qpn, priv->tx_ring_num);
1978 		priv->base_tx_qpn = 0;
1979 	}
1980 }
1981 
1982 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1983 {
1984 	struct mlx4_en_port_profile *prof = priv->prof;
1985 	int i;
1986 	int err;
1987 	int node;
1988 
1989 	err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &priv->base_tx_qpn);
1990 	if (err) {
1991 		en_err(priv, "failed reserving range for TX rings\n");
1992 		return err;
1993 	}
1994 
1995 	/* Create tx Rings */
1996 	for (i = 0; i < priv->tx_ring_num; i++) {
1997 		node = cpu_to_node(i % num_online_cpus());
1998 		if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1999 				      prof->tx_ring_size, i, TX, node))
2000 			goto err;
2001 
2002 		if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
2003 					   priv->base_tx_qpn + i,
2004 					   prof->tx_ring_size, TXBB_SIZE,
2005 					   node, i))
2006 			goto err;
2007 	}
2008 
2009 	/* Create rx Rings */
2010 	for (i = 0; i < priv->rx_ring_num; i++) {
2011 		node = cpu_to_node(i % num_online_cpus());
2012 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2013 				      prof->rx_ring_size, i, RX, node))
2014 			goto err;
2015 
2016 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2017 					   prof->rx_ring_size, priv->stride,
2018 					   node))
2019 			goto err;
2020 	}
2021 
2022 #ifdef CONFIG_RFS_ACCEL
2023 	if (priv->mdev->dev->caps.comp_pool) {
2024 		priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->mdev->dev->caps.comp_pool);
2025 		if (!priv->dev->rx_cpu_rmap)
2026 			goto err;
2027 	}
2028 #endif
2029 
2030 	return 0;
2031 
2032 err:
2033 	en_err(priv, "Failed to allocate NIC resources\n");
2034 	for (i = 0; i < priv->rx_ring_num; i++) {
2035 		if (priv->rx_ring[i])
2036 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2037 						prof->rx_ring_size,
2038 						priv->stride);
2039 		if (priv->rx_cq[i])
2040 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2041 	}
2042 	for (i = 0; i < priv->tx_ring_num; i++) {
2043 		if (priv->tx_ring[i])
2044 			mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
2045 		if (priv->tx_cq[i])
2046 			mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
2047 	}
2048 	return -ENOMEM;
2049 }
2050 
2051 
2052 void mlx4_en_destroy_netdev(struct net_device *dev)
2053 {
2054 	struct mlx4_en_priv *priv = netdev_priv(dev);
2055 	struct mlx4_en_dev *mdev = priv->mdev;
2056 
2057 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2058 
2059 	/* Unregister device - this will close the port if it was up */
2060 	if (priv->registered)
2061 		unregister_netdev(dev);
2062 
2063 	if (priv->allocated)
2064 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2065 
2066 	cancel_delayed_work(&priv->stats_task);
2067 	cancel_delayed_work(&priv->service_task);
2068 	/* flush any pending task for this netdev */
2069 	flush_workqueue(mdev->workqueue);
2070 
2071 	/* Detach the netdev so tasks would not attempt to access it */
2072 	mutex_lock(&mdev->state_lock);
2073 	mdev->pndev[priv->port] = NULL;
2074 	mutex_unlock(&mdev->state_lock);
2075 
2076 	mlx4_en_free_resources(priv);
2077 
2078 	kfree(priv->tx_ring);
2079 	kfree(priv->tx_cq);
2080 
2081 	free_netdev(dev);
2082 }
2083 
2084 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2085 {
2086 	struct mlx4_en_priv *priv = netdev_priv(dev);
2087 	struct mlx4_en_dev *mdev = priv->mdev;
2088 	int err = 0;
2089 
2090 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2091 		 dev->mtu, new_mtu);
2092 
2093 	if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
2094 		en_err(priv, "Bad MTU size:%d.\n", new_mtu);
2095 		return -EPERM;
2096 	}
2097 	dev->mtu = new_mtu;
2098 
2099 	if (netif_running(dev)) {
2100 		mutex_lock(&mdev->state_lock);
2101 		if (!mdev->device_up) {
2102 			/* NIC is probably restarting - let watchdog task reset
2103 			 * the port */
2104 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2105 		} else {
2106 			mlx4_en_stop_port(dev, 1);
2107 			err = mlx4_en_start_port(dev);
2108 			if (err) {
2109 				en_err(priv, "Failed restarting port:%d\n",
2110 					 priv->port);
2111 				queue_work(mdev->workqueue, &priv->watchdog_task);
2112 			}
2113 		}
2114 		mutex_unlock(&mdev->state_lock);
2115 	}
2116 	return 0;
2117 }
2118 
2119 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2120 {
2121 	struct mlx4_en_priv *priv = netdev_priv(dev);
2122 	struct mlx4_en_dev *mdev = priv->mdev;
2123 	struct hwtstamp_config config;
2124 
2125 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2126 		return -EFAULT;
2127 
2128 	/* reserved for future extensions */
2129 	if (config.flags)
2130 		return -EINVAL;
2131 
2132 	/* device doesn't support time stamping */
2133 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2134 		return -EINVAL;
2135 
2136 	/* TX HW timestamp */
2137 	switch (config.tx_type) {
2138 	case HWTSTAMP_TX_OFF:
2139 	case HWTSTAMP_TX_ON:
2140 		break;
2141 	default:
2142 		return -ERANGE;
2143 	}
2144 
2145 	/* RX HW timestamp */
2146 	switch (config.rx_filter) {
2147 	case HWTSTAMP_FILTER_NONE:
2148 		break;
2149 	case HWTSTAMP_FILTER_ALL:
2150 	case HWTSTAMP_FILTER_SOME:
2151 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2152 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2153 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2154 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2155 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2156 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2157 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2158 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2159 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2160 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2161 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2162 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2163 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2164 		break;
2165 	default:
2166 		return -ERANGE;
2167 	}
2168 
2169 	if (mlx4_en_timestamp_config(dev, config.tx_type, config.rx_filter)) {
2170 		config.tx_type = HWTSTAMP_TX_OFF;
2171 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2172 	}
2173 
2174 	return copy_to_user(ifr->ifr_data, &config,
2175 			    sizeof(config)) ? -EFAULT : 0;
2176 }
2177 
2178 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2179 {
2180 	struct mlx4_en_priv *priv = netdev_priv(dev);
2181 
2182 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2183 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2184 }
2185 
2186 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2187 {
2188 	switch (cmd) {
2189 	case SIOCSHWTSTAMP:
2190 		return mlx4_en_hwtstamp_set(dev, ifr);
2191 	case SIOCGHWTSTAMP:
2192 		return mlx4_en_hwtstamp_get(dev, ifr);
2193 	default:
2194 		return -EOPNOTSUPP;
2195 	}
2196 }
2197 
2198 static int mlx4_en_set_features(struct net_device *netdev,
2199 		netdev_features_t features)
2200 {
2201 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2202 
2203 	if (features & NETIF_F_LOOPBACK)
2204 		priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
2205 	else
2206 		priv->ctrl_flags &=
2207 			cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
2208 
2209 	mlx4_en_update_loopback_state(netdev, features);
2210 
2211 	return 0;
2212 
2213 }
2214 
2215 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2216 {
2217 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2218 	struct mlx4_en_dev *mdev = en_priv->mdev;
2219 	u64 mac_u64 = mlx4_en_mac_to_u64(mac);
2220 
2221 	if (!is_valid_ether_addr(mac))
2222 		return -EINVAL;
2223 
2224 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2225 }
2226 
2227 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos)
2228 {
2229 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2230 	struct mlx4_en_dev *mdev = en_priv->mdev;
2231 
2232 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos);
2233 }
2234 
2235 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2236 {
2237 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2238 	struct mlx4_en_dev *mdev = en_priv->mdev;
2239 
2240 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2241 }
2242 
2243 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2244 {
2245 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2246 	struct mlx4_en_dev *mdev = en_priv->mdev;
2247 
2248 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2249 }
2250 
2251 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2252 {
2253 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2254 	struct mlx4_en_dev *mdev = en_priv->mdev;
2255 
2256 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2257 }
2258 
2259 #define PORT_ID_BYTE_LEN 8
2260 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2261 				    struct netdev_phys_port_id *ppid)
2262 {
2263 	struct mlx4_en_priv *priv = netdev_priv(dev);
2264 	struct mlx4_dev *mdev = priv->mdev->dev;
2265 	int i;
2266 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2267 
2268 	if (!phys_port_id)
2269 		return -EOPNOTSUPP;
2270 
2271 	ppid->id_len = sizeof(phys_port_id);
2272 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2273 		ppid->id[i] =  phys_port_id & 0xff;
2274 		phys_port_id >>= 8;
2275 	}
2276 	return 0;
2277 }
2278 
2279 static const struct net_device_ops mlx4_netdev_ops = {
2280 	.ndo_open		= mlx4_en_open,
2281 	.ndo_stop		= mlx4_en_close,
2282 	.ndo_start_xmit		= mlx4_en_xmit,
2283 	.ndo_select_queue	= mlx4_en_select_queue,
2284 	.ndo_get_stats		= mlx4_en_get_stats,
2285 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2286 	.ndo_set_mac_address	= mlx4_en_set_mac,
2287 	.ndo_validate_addr	= eth_validate_addr,
2288 	.ndo_change_mtu		= mlx4_en_change_mtu,
2289 	.ndo_do_ioctl		= mlx4_en_ioctl,
2290 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2291 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2292 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2293 #ifdef CONFIG_NET_POLL_CONTROLLER
2294 	.ndo_poll_controller	= mlx4_en_netpoll,
2295 #endif
2296 	.ndo_set_features	= mlx4_en_set_features,
2297 	.ndo_setup_tc		= mlx4_en_setup_tc,
2298 #ifdef CONFIG_RFS_ACCEL
2299 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2300 #endif
2301 #ifdef CONFIG_NET_RX_BUSY_POLL
2302 	.ndo_busy_poll		= mlx4_en_low_latency_recv,
2303 #endif
2304 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2305 };
2306 
2307 static const struct net_device_ops mlx4_netdev_ops_master = {
2308 	.ndo_open		= mlx4_en_open,
2309 	.ndo_stop		= mlx4_en_close,
2310 	.ndo_start_xmit		= mlx4_en_xmit,
2311 	.ndo_select_queue	= mlx4_en_select_queue,
2312 	.ndo_get_stats		= mlx4_en_get_stats,
2313 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2314 	.ndo_set_mac_address	= mlx4_en_set_mac,
2315 	.ndo_validate_addr	= eth_validate_addr,
2316 	.ndo_change_mtu		= mlx4_en_change_mtu,
2317 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2318 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2319 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2320 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2321 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2322 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2323 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2324 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2325 #ifdef CONFIG_NET_POLL_CONTROLLER
2326 	.ndo_poll_controller	= mlx4_en_netpoll,
2327 #endif
2328 	.ndo_set_features	= mlx4_en_set_features,
2329 	.ndo_setup_tc		= mlx4_en_setup_tc,
2330 #ifdef CONFIG_RFS_ACCEL
2331 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2332 #endif
2333 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2334 };
2335 
2336 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2337 			struct mlx4_en_port_profile *prof)
2338 {
2339 	struct net_device *dev;
2340 	struct mlx4_en_priv *priv;
2341 	int i;
2342 	int err;
2343 	u64 mac_u64;
2344 
2345 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2346 				 MAX_TX_RINGS, MAX_RX_RINGS);
2347 	if (dev == NULL)
2348 		return -ENOMEM;
2349 
2350 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2351 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2352 
2353 	SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
2354 	dev->dev_id =  port - 1;
2355 
2356 	/*
2357 	 * Initialize driver private data
2358 	 */
2359 
2360 	priv = netdev_priv(dev);
2361 	memset(priv, 0, sizeof(struct mlx4_en_priv));
2362 	priv->dev = dev;
2363 	priv->mdev = mdev;
2364 	priv->ddev = &mdev->pdev->dev;
2365 	priv->prof = prof;
2366 	priv->port = port;
2367 	priv->port_up = false;
2368 	priv->flags = prof->flags;
2369 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2370 			MLX4_WQE_CTRL_SOLICITED);
2371 	priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2372 	priv->tx_ring_num = prof->tx_ring_num;
2373 
2374 	priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2375 				GFP_KERNEL);
2376 	if (!priv->tx_ring) {
2377 		err = -ENOMEM;
2378 		goto out;
2379 	}
2380 	priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2381 			      GFP_KERNEL);
2382 	if (!priv->tx_cq) {
2383 		err = -ENOMEM;
2384 		goto out;
2385 	}
2386 	priv->rx_ring_num = prof->rx_ring_num;
2387 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2388 	priv->mac_index = -1;
2389 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
2390 	spin_lock_init(&priv->stats_lock);
2391 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2392 	INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2393 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2394 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2395 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2396 #ifdef CONFIG_MLX4_EN_DCB
2397 	if (!mlx4_is_slave(priv->mdev->dev)) {
2398 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
2399 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2400 		} else {
2401 			en_info(priv, "enabling only PFC DCB ops\n");
2402 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2403 		}
2404 	}
2405 #endif
2406 
2407 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2408 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
2409 
2410 	/* Query for default mac and max mtu */
2411 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2412 
2413 	/* Set default MAC */
2414 	dev->addr_len = ETH_ALEN;
2415 	mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2416 	if (!is_valid_ether_addr(dev->dev_addr)) {
2417 		if (mlx4_is_slave(priv->mdev->dev)) {
2418 			eth_hw_addr_random(dev);
2419 			en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
2420 			mac_u64 = mlx4_en_mac_to_u64(dev->dev_addr);
2421 			mdev->dev->caps.def_mac[priv->port] = mac_u64;
2422 		} else {
2423 			en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2424 			       priv->port, dev->dev_addr);
2425 			err = -EINVAL;
2426 			goto out;
2427 		}
2428 	}
2429 
2430 	memcpy(priv->prev_mac, dev->dev_addr, sizeof(priv->prev_mac));
2431 
2432 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2433 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2434 	err = mlx4_en_alloc_resources(priv);
2435 	if (err)
2436 		goto out;
2437 
2438 #ifdef CONFIG_RFS_ACCEL
2439 	INIT_LIST_HEAD(&priv->filters);
2440 	spin_lock_init(&priv->filters_lock);
2441 #endif
2442 
2443 	/* Initialize time stamping config */
2444 	priv->hwtstamp_config.flags = 0;
2445 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
2446 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2447 
2448 	/* Allocate page for receive rings */
2449 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2450 				MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2451 	if (err) {
2452 		en_err(priv, "Failed to allocate page for rx qps\n");
2453 		goto out;
2454 	}
2455 	priv->allocated = 1;
2456 
2457 	/*
2458 	 * Initialize netdev entry points
2459 	 */
2460 	if (mlx4_is_master(priv->mdev->dev))
2461 		dev->netdev_ops = &mlx4_netdev_ops_master;
2462 	else
2463 		dev->netdev_ops = &mlx4_netdev_ops;
2464 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2465 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2466 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2467 
2468 	SET_ETHTOOL_OPS(dev, &mlx4_en_ethtool_ops);
2469 
2470 	/*
2471 	 * Set driver features
2472 	 */
2473 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2474 	if (mdev->LSO_support)
2475 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2476 
2477 	dev->vlan_features = dev->hw_features;
2478 
2479 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2480 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2481 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2482 			NETIF_F_HW_VLAN_CTAG_FILTER;
2483 	dev->hw_features |= NETIF_F_LOOPBACK;
2484 
2485 	if (mdev->dev->caps.steering_mode ==
2486 	    MLX4_STEERING_MODE_DEVICE_MANAGED)
2487 		dev->hw_features |= NETIF_F_NTUPLE;
2488 
2489 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
2490 		dev->priv_flags |= IFF_UNICAST_FLT;
2491 
2492 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2493 		dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
2494 					NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL;
2495 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
2496 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL;
2497 	}
2498 
2499 	mdev->pndev[port] = dev;
2500 
2501 	netif_carrier_off(dev);
2502 	mlx4_en_set_default_moderation(priv);
2503 
2504 	err = register_netdev(dev);
2505 	if (err) {
2506 		en_err(priv, "Netdev registration failed for port %d\n", port);
2507 		goto out;
2508 	}
2509 	priv->registered = 1;
2510 
2511 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2512 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2513 
2514 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
2515 
2516 	/* Configure port */
2517 	mlx4_en_calc_rx_buf(dev);
2518 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2519 				    priv->rx_skb_size + ETH_FCS_LEN,
2520 				    prof->tx_pause, prof->tx_ppp,
2521 				    prof->rx_pause, prof->rx_ppp);
2522 	if (err) {
2523 		en_err(priv, "Failed setting port general configurations "
2524 		       "for port %d, with error %d\n", priv->port, err);
2525 		goto out;
2526 	}
2527 
2528 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2529 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC);
2530 		if (err) {
2531 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
2532 			       err);
2533 			goto out;
2534 		}
2535 	}
2536 
2537 	/* Init port */
2538 	en_warn(priv, "Initializing port\n");
2539 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
2540 	if (err) {
2541 		en_err(priv, "Failed Initializing port\n");
2542 		goto out;
2543 	}
2544 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2545 
2546 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2547 		queue_delayed_work(mdev->workqueue, &priv->service_task,
2548 				   SERVICE_TASK_DELAY);
2549 
2550 	return 0;
2551 
2552 out:
2553 	mlx4_en_destroy_netdev(dev);
2554 	return err;
2555 }
2556 
2557