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 				return err;
746 			}
747 		}
748 		return -EINVAL;
749 	}
750 
751 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
752 }
753 
754 u64 mlx4_en_mac_to_u64(u8 *addr)
755 {
756 	u64 mac = 0;
757 	int i;
758 
759 	for (i = 0; i < ETH_ALEN; i++) {
760 		mac <<= 8;
761 		mac |= addr[i];
762 	}
763 	return mac;
764 }
765 
766 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv)
767 {
768 	int err = 0;
769 
770 	if (priv->port_up) {
771 		/* Remove old MAC and insert the new one */
772 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
773 					  priv->dev->dev_addr, priv->prev_mac);
774 		if (err)
775 			en_err(priv, "Failed changing HW MAC address\n");
776 		memcpy(priv->prev_mac, priv->dev->dev_addr,
777 		       sizeof(priv->prev_mac));
778 	} else
779 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
780 
781 	return err;
782 }
783 
784 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
785 {
786 	struct mlx4_en_priv *priv = netdev_priv(dev);
787 	struct mlx4_en_dev *mdev = priv->mdev;
788 	struct sockaddr *saddr = addr;
789 	int err;
790 
791 	if (!is_valid_ether_addr(saddr->sa_data))
792 		return -EADDRNOTAVAIL;
793 
794 	memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
795 
796 	mutex_lock(&mdev->state_lock);
797 	err = mlx4_en_do_set_mac(priv);
798 	mutex_unlock(&mdev->state_lock);
799 
800 	return err;
801 }
802 
803 static void mlx4_en_clear_list(struct net_device *dev)
804 {
805 	struct mlx4_en_priv *priv = netdev_priv(dev);
806 	struct mlx4_en_mc_list *tmp, *mc_to_del;
807 
808 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
809 		list_del(&mc_to_del->list);
810 		kfree(mc_to_del);
811 	}
812 }
813 
814 static void mlx4_en_cache_mclist(struct net_device *dev)
815 {
816 	struct mlx4_en_priv *priv = netdev_priv(dev);
817 	struct netdev_hw_addr *ha;
818 	struct mlx4_en_mc_list *tmp;
819 
820 	mlx4_en_clear_list(dev);
821 	netdev_for_each_mc_addr(ha, dev) {
822 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
823 		if (!tmp) {
824 			mlx4_en_clear_list(dev);
825 			return;
826 		}
827 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
828 		list_add_tail(&tmp->list, &priv->mc_list);
829 	}
830 }
831 
832 static void update_mclist_flags(struct mlx4_en_priv *priv,
833 				struct list_head *dst,
834 				struct list_head *src)
835 {
836 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
837 	bool found;
838 
839 	/* Find all the entries that should be removed from dst,
840 	 * These are the entries that are not found in src
841 	 */
842 	list_for_each_entry(dst_tmp, dst, list) {
843 		found = false;
844 		list_for_each_entry(src_tmp, src, list) {
845 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
846 				found = true;
847 				break;
848 			}
849 		}
850 		if (!found)
851 			dst_tmp->action = MCLIST_REM;
852 	}
853 
854 	/* Add entries that exist in src but not in dst
855 	 * mark them as need to add
856 	 */
857 	list_for_each_entry(src_tmp, src, list) {
858 		found = false;
859 		list_for_each_entry(dst_tmp, dst, list) {
860 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
861 				dst_tmp->action = MCLIST_NONE;
862 				found = true;
863 				break;
864 			}
865 		}
866 		if (!found) {
867 			new_mc = kmemdup(src_tmp,
868 					 sizeof(struct mlx4_en_mc_list),
869 					 GFP_KERNEL);
870 			if (!new_mc)
871 				return;
872 
873 			new_mc->action = MCLIST_ADD;
874 			list_add_tail(&new_mc->list, dst);
875 		}
876 	}
877 }
878 
879 static void mlx4_en_set_rx_mode(struct net_device *dev)
880 {
881 	struct mlx4_en_priv *priv = netdev_priv(dev);
882 
883 	if (!priv->port_up)
884 		return;
885 
886 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
887 }
888 
889 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
890 				     struct mlx4_en_dev *mdev)
891 {
892 	int err = 0;
893 
894 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
895 		if (netif_msg_rx_status(priv))
896 			en_warn(priv, "Entering promiscuous mode\n");
897 		priv->flags |= MLX4_EN_FLAG_PROMISC;
898 
899 		/* Enable promiscouos mode */
900 		switch (mdev->dev->caps.steering_mode) {
901 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
902 			err = mlx4_flow_steer_promisc_add(mdev->dev,
903 							  priv->port,
904 							  priv->base_qpn,
905 							  MLX4_FS_ALL_DEFAULT);
906 			if (err)
907 				en_err(priv, "Failed enabling promiscuous mode\n");
908 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
909 			break;
910 
911 		case MLX4_STEERING_MODE_B0:
912 			err = mlx4_unicast_promisc_add(mdev->dev,
913 						       priv->base_qpn,
914 						       priv->port);
915 			if (err)
916 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
917 
918 			/* Add the default qp number as multicast
919 			 * promisc
920 			 */
921 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
922 				err = mlx4_multicast_promisc_add(mdev->dev,
923 								 priv->base_qpn,
924 								 priv->port);
925 				if (err)
926 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
927 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
928 			}
929 			break;
930 
931 		case MLX4_STEERING_MODE_A0:
932 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
933 						     priv->port,
934 						     priv->base_qpn,
935 						     1);
936 			if (err)
937 				en_err(priv, "Failed enabling promiscuous mode\n");
938 			break;
939 		}
940 
941 		/* Disable port multicast filter (unconditionally) */
942 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
943 					  0, MLX4_MCAST_DISABLE);
944 		if (err)
945 			en_err(priv, "Failed disabling multicast filter\n");
946 
947 		/* Disable port VLAN filter */
948 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
949 		if (err)
950 			en_err(priv, "Failed disabling VLAN filter\n");
951 	}
952 }
953 
954 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
955 				       struct mlx4_en_dev *mdev)
956 {
957 	int err = 0;
958 
959 	if (netif_msg_rx_status(priv))
960 		en_warn(priv, "Leaving promiscuous mode\n");
961 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
962 
963 	/* Disable promiscouos mode */
964 	switch (mdev->dev->caps.steering_mode) {
965 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
966 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
967 						     priv->port,
968 						     MLX4_FS_ALL_DEFAULT);
969 		if (err)
970 			en_err(priv, "Failed disabling promiscuous mode\n");
971 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
972 		break;
973 
974 	case MLX4_STEERING_MODE_B0:
975 		err = mlx4_unicast_promisc_remove(mdev->dev,
976 						  priv->base_qpn,
977 						  priv->port);
978 		if (err)
979 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
980 		/* Disable Multicast promisc */
981 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
982 			err = mlx4_multicast_promisc_remove(mdev->dev,
983 							    priv->base_qpn,
984 							    priv->port);
985 			if (err)
986 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
987 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
988 		}
989 		break;
990 
991 	case MLX4_STEERING_MODE_A0:
992 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
993 					     priv->port,
994 					     priv->base_qpn, 0);
995 		if (err)
996 			en_err(priv, "Failed disabling promiscuous mode\n");
997 		break;
998 	}
999 
1000 	/* Enable port VLAN filter */
1001 	err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
1002 	if (err)
1003 		en_err(priv, "Failed enabling VLAN filter\n");
1004 }
1005 
1006 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
1007 				 struct net_device *dev,
1008 				 struct mlx4_en_dev *mdev)
1009 {
1010 	struct mlx4_en_mc_list *mclist, *tmp;
1011 	u64 mcast_addr = 0;
1012 	u8 mc_list[16] = {0};
1013 	int err = 0;
1014 
1015 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
1016 	if (dev->flags & IFF_ALLMULTI) {
1017 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1018 					  0, MLX4_MCAST_DISABLE);
1019 		if (err)
1020 			en_err(priv, "Failed disabling multicast filter\n");
1021 
1022 		/* Add the default qp number as multicast promisc */
1023 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1024 			switch (mdev->dev->caps.steering_mode) {
1025 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1026 				err = mlx4_flow_steer_promisc_add(mdev->dev,
1027 								  priv->port,
1028 								  priv->base_qpn,
1029 								  MLX4_FS_MC_DEFAULT);
1030 				break;
1031 
1032 			case MLX4_STEERING_MODE_B0:
1033 				err = mlx4_multicast_promisc_add(mdev->dev,
1034 								 priv->base_qpn,
1035 								 priv->port);
1036 				break;
1037 
1038 			case MLX4_STEERING_MODE_A0:
1039 				break;
1040 			}
1041 			if (err)
1042 				en_err(priv, "Failed entering multicast promisc mode\n");
1043 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1044 		}
1045 	} else {
1046 		/* Disable Multicast promisc */
1047 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1048 			switch (mdev->dev->caps.steering_mode) {
1049 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1050 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
1051 								     priv->port,
1052 								     MLX4_FS_MC_DEFAULT);
1053 				break;
1054 
1055 			case MLX4_STEERING_MODE_B0:
1056 				err = mlx4_multicast_promisc_remove(mdev->dev,
1057 								    priv->base_qpn,
1058 								    priv->port);
1059 				break;
1060 
1061 			case MLX4_STEERING_MODE_A0:
1062 				break;
1063 			}
1064 			if (err)
1065 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
1066 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1067 		}
1068 
1069 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1070 					  0, MLX4_MCAST_DISABLE);
1071 		if (err)
1072 			en_err(priv, "Failed disabling multicast filter\n");
1073 
1074 		/* Flush mcast filter and init it with broadcast address */
1075 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1076 				    1, MLX4_MCAST_CONFIG);
1077 
1078 		/* Update multicast list - we cache all addresses so they won't
1079 		 * change while HW is updated holding the command semaphor */
1080 		netif_addr_lock_bh(dev);
1081 		mlx4_en_cache_mclist(dev);
1082 		netif_addr_unlock_bh(dev);
1083 		list_for_each_entry(mclist, &priv->mc_list, list) {
1084 			mcast_addr = mlx4_en_mac_to_u64(mclist->addr);
1085 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1086 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
1087 		}
1088 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1089 					  0, MLX4_MCAST_ENABLE);
1090 		if (err)
1091 			en_err(priv, "Failed enabling multicast filter\n");
1092 
1093 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1094 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1095 			if (mclist->action == MCLIST_REM) {
1096 				/* detach this address and delete from list */
1097 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1098 				mc_list[5] = priv->port;
1099 				err = mlx4_multicast_detach(mdev->dev,
1100 							    &priv->rss_map.indir_qp,
1101 							    mc_list,
1102 							    MLX4_PROT_ETH,
1103 							    mclist->reg_id);
1104 				if (err)
1105 					en_err(priv, "Fail to detach multicast address\n");
1106 
1107 				if (mclist->tunnel_reg_id) {
1108 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1109 					if (err)
1110 						en_err(priv, "Failed to detach multicast address\n");
1111 				}
1112 
1113 				/* remove from list */
1114 				list_del(&mclist->list);
1115 				kfree(mclist);
1116 			} else if (mclist->action == MCLIST_ADD) {
1117 				/* attach the address */
1118 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1119 				/* needed for B0 steering support */
1120 				mc_list[5] = priv->port;
1121 				err = mlx4_multicast_attach(mdev->dev,
1122 							    &priv->rss_map.indir_qp,
1123 							    mc_list,
1124 							    priv->port, 0,
1125 							    MLX4_PROT_ETH,
1126 							    &mclist->reg_id);
1127 				if (err)
1128 					en_err(priv, "Fail to attach multicast address\n");
1129 
1130 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1131 							       &mclist->tunnel_reg_id);
1132 				if (err)
1133 					en_err(priv, "Failed to attach multicast address\n");
1134 			}
1135 		}
1136 	}
1137 }
1138 
1139 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1140 				 struct net_device *dev,
1141 				 struct mlx4_en_dev *mdev)
1142 {
1143 	struct netdev_hw_addr *ha;
1144 	struct mlx4_mac_entry *entry;
1145 	struct hlist_node *tmp;
1146 	bool found;
1147 	u64 mac;
1148 	int err = 0;
1149 	struct hlist_head *bucket;
1150 	unsigned int i;
1151 	int removed = 0;
1152 	u32 prev_flags;
1153 
1154 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1155 	 * since all modification code is protected by mdev->state_lock
1156 	 */
1157 
1158 	/* find what to remove */
1159 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1160 		bucket = &priv->mac_hash[i];
1161 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1162 			found = false;
1163 			netdev_for_each_uc_addr(ha, dev) {
1164 				if (ether_addr_equal_64bits(entry->mac,
1165 							    ha->addr)) {
1166 					found = true;
1167 					break;
1168 				}
1169 			}
1170 
1171 			/* MAC address of the port is not in uc list */
1172 			if (ether_addr_equal_64bits(entry->mac, dev->dev_addr))
1173 				found = true;
1174 
1175 			if (!found) {
1176 				mac = mlx4_en_mac_to_u64(entry->mac);
1177 				mlx4_en_uc_steer_release(priv, entry->mac,
1178 							 priv->base_qpn,
1179 							 entry->reg_id);
1180 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1181 
1182 				hlist_del_rcu(&entry->hlist);
1183 				kfree_rcu(entry, rcu);
1184 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1185 				       entry->mac, priv->port);
1186 				++removed;
1187 			}
1188 		}
1189 	}
1190 
1191 	/* if we didn't remove anything, there is no use in trying to add
1192 	 * again once we are in a forced promisc mode state
1193 	 */
1194 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1195 		return;
1196 
1197 	prev_flags = priv->flags;
1198 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1199 
1200 	/* find what to add */
1201 	netdev_for_each_uc_addr(ha, dev) {
1202 		found = false;
1203 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1204 		hlist_for_each_entry(entry, bucket, hlist) {
1205 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1206 				found = true;
1207 				break;
1208 			}
1209 		}
1210 
1211 		if (!found) {
1212 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1213 			if (!entry) {
1214 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1215 				       ha->addr, priv->port);
1216 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1217 				break;
1218 			}
1219 			mac = mlx4_en_mac_to_u64(ha->addr);
1220 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1221 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1222 			if (err < 0) {
1223 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1224 				       ha->addr, priv->port, err);
1225 				kfree(entry);
1226 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1227 				break;
1228 			}
1229 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1230 						   &priv->base_qpn,
1231 						   &entry->reg_id);
1232 			if (err) {
1233 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1234 				       ha->addr, priv->port, err);
1235 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1236 				kfree(entry);
1237 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1238 				break;
1239 			} else {
1240 				unsigned int mac_hash;
1241 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1242 				       ha->addr, priv->port);
1243 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1244 				bucket = &priv->mac_hash[mac_hash];
1245 				hlist_add_head_rcu(&entry->hlist, bucket);
1246 			}
1247 		}
1248 	}
1249 
1250 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1251 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1252 			priv->port);
1253 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1254 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1255 			priv->port);
1256 	}
1257 }
1258 
1259 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1260 {
1261 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1262 						 rx_mode_task);
1263 	struct mlx4_en_dev *mdev = priv->mdev;
1264 	struct net_device *dev = priv->dev;
1265 
1266 	mutex_lock(&mdev->state_lock);
1267 	if (!mdev->device_up) {
1268 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1269 		goto out;
1270 	}
1271 	if (!priv->port_up) {
1272 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1273 		goto out;
1274 	}
1275 
1276 	if (!netif_carrier_ok(dev)) {
1277 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1278 			if (priv->port_state.link_state) {
1279 				priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1280 				netif_carrier_on(dev);
1281 				en_dbg(LINK, priv, "Link Up\n");
1282 			}
1283 		}
1284 	}
1285 
1286 	if (dev->priv_flags & IFF_UNICAST_FLT)
1287 		mlx4_en_do_uc_filter(priv, dev, mdev);
1288 
1289 	/* Promsicuous mode: disable all filters */
1290 	if ((dev->flags & IFF_PROMISC) ||
1291 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1292 		mlx4_en_set_promisc_mode(priv, mdev);
1293 		goto out;
1294 	}
1295 
1296 	/* Not in promiscuous mode */
1297 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1298 		mlx4_en_clear_promisc_mode(priv, mdev);
1299 
1300 	mlx4_en_do_multicast(priv, dev, mdev);
1301 out:
1302 	mutex_unlock(&mdev->state_lock);
1303 }
1304 
1305 #ifdef CONFIG_NET_POLL_CONTROLLER
1306 static void mlx4_en_netpoll(struct net_device *dev)
1307 {
1308 	struct mlx4_en_priv *priv = netdev_priv(dev);
1309 	struct mlx4_en_cq *cq;
1310 	unsigned long flags;
1311 	int i;
1312 
1313 	for (i = 0; i < priv->rx_ring_num; i++) {
1314 		cq = priv->rx_cq[i];
1315 		spin_lock_irqsave(&cq->lock, flags);
1316 		napi_synchronize(&cq->napi);
1317 		mlx4_en_process_rx_cq(dev, cq, 0);
1318 		spin_unlock_irqrestore(&cq->lock, flags);
1319 	}
1320 }
1321 #endif
1322 
1323 static void mlx4_en_tx_timeout(struct net_device *dev)
1324 {
1325 	struct mlx4_en_priv *priv = netdev_priv(dev);
1326 	struct mlx4_en_dev *mdev = priv->mdev;
1327 	int i;
1328 
1329 	if (netif_msg_timer(priv))
1330 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1331 
1332 	for (i = 0; i < priv->tx_ring_num; i++) {
1333 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1334 			continue;
1335 		en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1336 			i, priv->tx_ring[i]->qpn, priv->tx_ring[i]->cqn,
1337 			priv->tx_ring[i]->cons, priv->tx_ring[i]->prod);
1338 	}
1339 
1340 	priv->port_stats.tx_timeout++;
1341 	en_dbg(DRV, priv, "Scheduling watchdog\n");
1342 	queue_work(mdev->workqueue, &priv->watchdog_task);
1343 }
1344 
1345 
1346 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
1347 {
1348 	struct mlx4_en_priv *priv = netdev_priv(dev);
1349 
1350 	spin_lock_bh(&priv->stats_lock);
1351 	memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
1352 	spin_unlock_bh(&priv->stats_lock);
1353 
1354 	return &priv->ret_stats;
1355 }
1356 
1357 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1358 {
1359 	struct mlx4_en_cq *cq;
1360 	int i;
1361 
1362 	/* If we haven't received a specific coalescing setting
1363 	 * (module param), we set the moderation parameters as follows:
1364 	 * - moder_cnt is set to the number of mtu sized packets to
1365 	 *   satisfy our coalescing target.
1366 	 * - moder_time is set to a fixed value.
1367 	 */
1368 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1369 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1370 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1371 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1372 	en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1373 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1374 
1375 	/* Setup cq moderation params */
1376 	for (i = 0; i < priv->rx_ring_num; i++) {
1377 		cq = priv->rx_cq[i];
1378 		cq->moder_cnt = priv->rx_frames;
1379 		cq->moder_time = priv->rx_usecs;
1380 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1381 		priv->last_moder_packets[i] = 0;
1382 		priv->last_moder_bytes[i] = 0;
1383 	}
1384 
1385 	for (i = 0; i < priv->tx_ring_num; i++) {
1386 		cq = priv->tx_cq[i];
1387 		cq->moder_cnt = priv->tx_frames;
1388 		cq->moder_time = priv->tx_usecs;
1389 	}
1390 
1391 	/* Reset auto-moderation params */
1392 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1393 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1394 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1395 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1396 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1397 	priv->adaptive_rx_coal = 1;
1398 	priv->last_moder_jiffies = 0;
1399 	priv->last_moder_tx_packets = 0;
1400 }
1401 
1402 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1403 {
1404 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1405 	struct mlx4_en_cq *cq;
1406 	unsigned long packets;
1407 	unsigned long rate;
1408 	unsigned long avg_pkt_size;
1409 	unsigned long rx_packets;
1410 	unsigned long rx_bytes;
1411 	unsigned long rx_pkt_diff;
1412 	int moder_time;
1413 	int ring, err;
1414 
1415 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1416 		return;
1417 
1418 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1419 		spin_lock_bh(&priv->stats_lock);
1420 		rx_packets = priv->rx_ring[ring]->packets;
1421 		rx_bytes = priv->rx_ring[ring]->bytes;
1422 		spin_unlock_bh(&priv->stats_lock);
1423 
1424 		rx_pkt_diff = ((unsigned long) (rx_packets -
1425 				priv->last_moder_packets[ring]));
1426 		packets = rx_pkt_diff;
1427 		rate = packets * HZ / period;
1428 		avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1429 				priv->last_moder_bytes[ring])) / packets : 0;
1430 
1431 		/* Apply auto-moderation only when packet rate
1432 		 * exceeds a rate that it matters */
1433 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1434 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1435 			if (rate < priv->pkt_rate_low)
1436 				moder_time = priv->rx_usecs_low;
1437 			else if (rate > priv->pkt_rate_high)
1438 				moder_time = priv->rx_usecs_high;
1439 			else
1440 				moder_time = (rate - priv->pkt_rate_low) *
1441 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1442 					(priv->pkt_rate_high - priv->pkt_rate_low) +
1443 					priv->rx_usecs_low;
1444 		} else {
1445 			moder_time = priv->rx_usecs_low;
1446 		}
1447 
1448 		if (moder_time != priv->last_moder_time[ring]) {
1449 			priv->last_moder_time[ring] = moder_time;
1450 			cq = priv->rx_cq[ring];
1451 			cq->moder_time = moder_time;
1452 			cq->moder_cnt = priv->rx_frames;
1453 			err = mlx4_en_set_cq_moder(priv, cq);
1454 			if (err)
1455 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1456 				       ring);
1457 		}
1458 		priv->last_moder_packets[ring] = rx_packets;
1459 		priv->last_moder_bytes[ring] = rx_bytes;
1460 	}
1461 
1462 	priv->last_moder_jiffies = jiffies;
1463 }
1464 
1465 static void mlx4_en_do_get_stats(struct work_struct *work)
1466 {
1467 	struct delayed_work *delay = to_delayed_work(work);
1468 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1469 						 stats_task);
1470 	struct mlx4_en_dev *mdev = priv->mdev;
1471 	int err;
1472 
1473 	mutex_lock(&mdev->state_lock);
1474 	if (mdev->device_up) {
1475 		if (priv->port_up) {
1476 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1477 			if (err)
1478 				en_dbg(HW, priv, "Could not update stats\n");
1479 
1480 			mlx4_en_auto_moderation(priv);
1481 		}
1482 
1483 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1484 	}
1485 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1486 		mlx4_en_do_set_mac(priv);
1487 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1488 	}
1489 	mutex_unlock(&mdev->state_lock);
1490 }
1491 
1492 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1493  * periodically
1494  */
1495 static void mlx4_en_service_task(struct work_struct *work)
1496 {
1497 	struct delayed_work *delay = to_delayed_work(work);
1498 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1499 						 service_task);
1500 	struct mlx4_en_dev *mdev = priv->mdev;
1501 
1502 	mutex_lock(&mdev->state_lock);
1503 	if (mdev->device_up) {
1504 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1505 			mlx4_en_ptp_overflow_check(mdev);
1506 
1507 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1508 				   SERVICE_TASK_DELAY);
1509 	}
1510 	mutex_unlock(&mdev->state_lock);
1511 }
1512 
1513 static void mlx4_en_linkstate(struct work_struct *work)
1514 {
1515 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1516 						 linkstate_task);
1517 	struct mlx4_en_dev *mdev = priv->mdev;
1518 	int linkstate = priv->link_state;
1519 
1520 	mutex_lock(&mdev->state_lock);
1521 	/* If observable port state changed set carrier state and
1522 	 * report to system log */
1523 	if (priv->last_link_state != linkstate) {
1524 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1525 			en_info(priv, "Link Down\n");
1526 			netif_carrier_off(priv->dev);
1527 		} else {
1528 			en_info(priv, "Link Up\n");
1529 			netif_carrier_on(priv->dev);
1530 		}
1531 	}
1532 	priv->last_link_state = linkstate;
1533 	mutex_unlock(&mdev->state_lock);
1534 }
1535 
1536 
1537 int mlx4_en_start_port(struct net_device *dev)
1538 {
1539 	struct mlx4_en_priv *priv = netdev_priv(dev);
1540 	struct mlx4_en_dev *mdev = priv->mdev;
1541 	struct mlx4_en_cq *cq;
1542 	struct mlx4_en_tx_ring *tx_ring;
1543 	int rx_index = 0;
1544 	int tx_index = 0;
1545 	int err = 0;
1546 	int i;
1547 	int j;
1548 	u8 mc_list[16] = {0};
1549 
1550 	if (priv->port_up) {
1551 		en_dbg(DRV, priv, "start port called while port already up\n");
1552 		return 0;
1553 	}
1554 
1555 	INIT_LIST_HEAD(&priv->mc_list);
1556 	INIT_LIST_HEAD(&priv->curr_list);
1557 	INIT_LIST_HEAD(&priv->ethtool_list);
1558 	memset(&priv->ethtool_rules[0], 0,
1559 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1560 
1561 	/* Calculate Rx buf size */
1562 	dev->mtu = min(dev->mtu, priv->max_mtu);
1563 	mlx4_en_calc_rx_buf(dev);
1564 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1565 
1566 	/* Configure rx cq's and rings */
1567 	err = mlx4_en_activate_rx_rings(priv);
1568 	if (err) {
1569 		en_err(priv, "Failed to activate RX rings\n");
1570 		return err;
1571 	}
1572 	for (i = 0; i < priv->rx_ring_num; i++) {
1573 		cq = priv->rx_cq[i];
1574 
1575 		mlx4_en_cq_init_lock(cq);
1576 
1577 		err = mlx4_en_activate_cq(priv, cq, i);
1578 		if (err) {
1579 			en_err(priv, "Failed activating Rx CQ\n");
1580 			goto cq_err;
1581 		}
1582 		for (j = 0; j < cq->size; j++)
1583 			cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1584 		err = mlx4_en_set_cq_moder(priv, cq);
1585 		if (err) {
1586 			en_err(priv, "Failed setting cq moderation parameters");
1587 			mlx4_en_deactivate_cq(priv, cq);
1588 			goto cq_err;
1589 		}
1590 		mlx4_en_arm_cq(priv, cq);
1591 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1592 		++rx_index;
1593 	}
1594 
1595 	/* Set qp number */
1596 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1597 	err = mlx4_en_get_qp(priv);
1598 	if (err) {
1599 		en_err(priv, "Failed getting eth qp\n");
1600 		goto cq_err;
1601 	}
1602 	mdev->mac_removed[priv->port] = 0;
1603 
1604 	err = mlx4_en_config_rss_steer(priv);
1605 	if (err) {
1606 		en_err(priv, "Failed configuring rss steering\n");
1607 		goto mac_err;
1608 	}
1609 
1610 	err = mlx4_en_create_drop_qp(priv);
1611 	if (err)
1612 		goto rss_err;
1613 
1614 	/* Configure tx cq's and rings */
1615 	for (i = 0; i < priv->tx_ring_num; i++) {
1616 		/* Configure cq */
1617 		cq = priv->tx_cq[i];
1618 		err = mlx4_en_activate_cq(priv, cq, i);
1619 		if (err) {
1620 			en_err(priv, "Failed allocating Tx CQ\n");
1621 			goto tx_err;
1622 		}
1623 		err = mlx4_en_set_cq_moder(priv, cq);
1624 		if (err) {
1625 			en_err(priv, "Failed setting cq moderation parameters");
1626 			mlx4_en_deactivate_cq(priv, cq);
1627 			goto tx_err;
1628 		}
1629 		en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1630 		cq->buf->wqe_index = cpu_to_be16(0xffff);
1631 
1632 		/* Configure ring */
1633 		tx_ring = priv->tx_ring[i];
1634 		err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1635 			i / priv->num_tx_rings_p_up);
1636 		if (err) {
1637 			en_err(priv, "Failed allocating Tx ring\n");
1638 			mlx4_en_deactivate_cq(priv, cq);
1639 			goto tx_err;
1640 		}
1641 		tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1642 
1643 		/* Arm CQ for TX completions */
1644 		mlx4_en_arm_cq(priv, cq);
1645 
1646 		/* Set initial ownership of all Tx TXBBs to SW (1) */
1647 		for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1648 			*((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1649 		++tx_index;
1650 	}
1651 
1652 	/* Configure port */
1653 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1654 				    priv->rx_skb_size + ETH_FCS_LEN,
1655 				    priv->prof->tx_pause,
1656 				    priv->prof->tx_ppp,
1657 				    priv->prof->rx_pause,
1658 				    priv->prof->rx_ppp);
1659 	if (err) {
1660 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1661 		       priv->port, err);
1662 		goto tx_err;
1663 	}
1664 	/* Set default qp number */
1665 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1666 	if (err) {
1667 		en_err(priv, "Failed setting default qp numbers\n");
1668 		goto tx_err;
1669 	}
1670 
1671 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1672 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC);
1673 		if (err) {
1674 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1675 			       err);
1676 			goto tx_err;
1677 		}
1678 	}
1679 
1680 	/* Init port */
1681 	en_dbg(HW, priv, "Initializing port\n");
1682 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1683 	if (err) {
1684 		en_err(priv, "Failed Initializing port\n");
1685 		goto tx_err;
1686 	}
1687 
1688 	/* Attach rx QP to bradcast address */
1689 	memset(&mc_list[10], 0xff, ETH_ALEN);
1690 	mc_list[5] = priv->port; /* needed for B0 steering support */
1691 	if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1692 				  priv->port, 0, MLX4_PROT_ETH,
1693 				  &priv->broadcast_id))
1694 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1695 
1696 	/* Must redo promiscuous mode setup. */
1697 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1698 
1699 	/* Schedule multicast task to populate multicast list */
1700 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1701 
1702 	mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
1703 
1704 	priv->port_up = true;
1705 	netif_tx_start_all_queues(dev);
1706 	netif_device_attach(dev);
1707 
1708 	return 0;
1709 
1710 tx_err:
1711 	while (tx_index--) {
1712 		mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1713 		mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1714 	}
1715 	mlx4_en_destroy_drop_qp(priv);
1716 rss_err:
1717 	mlx4_en_release_rss_steer(priv);
1718 mac_err:
1719 	mlx4_en_put_qp(priv);
1720 cq_err:
1721 	while (rx_index--)
1722 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1723 	for (i = 0; i < priv->rx_ring_num; i++)
1724 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1725 
1726 	return err; /* need to close devices */
1727 }
1728 
1729 
1730 void mlx4_en_stop_port(struct net_device *dev, int detach)
1731 {
1732 	struct mlx4_en_priv *priv = netdev_priv(dev);
1733 	struct mlx4_en_dev *mdev = priv->mdev;
1734 	struct mlx4_en_mc_list *mclist, *tmp;
1735 	struct ethtool_flow_id *flow, *tmp_flow;
1736 	int i;
1737 	u8 mc_list[16] = {0};
1738 
1739 	if (!priv->port_up) {
1740 		en_dbg(DRV, priv, "stop port called while port already down\n");
1741 		return;
1742 	}
1743 
1744 	/* close port*/
1745 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1746 
1747 	/* Synchronize with tx routine */
1748 	netif_tx_lock_bh(dev);
1749 	if (detach)
1750 		netif_device_detach(dev);
1751 	netif_tx_stop_all_queues(dev);
1752 	netif_tx_unlock_bh(dev);
1753 
1754 	netif_tx_disable(dev);
1755 
1756 	/* Set port as not active */
1757 	priv->port_up = false;
1758 
1759 	/* Promsicuous mode */
1760 	if (mdev->dev->caps.steering_mode ==
1761 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1762 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1763 				 MLX4_EN_FLAG_MC_PROMISC);
1764 		mlx4_flow_steer_promisc_remove(mdev->dev,
1765 					       priv->port,
1766 					       MLX4_FS_ALL_DEFAULT);
1767 		mlx4_flow_steer_promisc_remove(mdev->dev,
1768 					       priv->port,
1769 					       MLX4_FS_MC_DEFAULT);
1770 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1771 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1772 
1773 		/* Disable promiscouos mode */
1774 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1775 					    priv->port);
1776 
1777 		/* Disable Multicast promisc */
1778 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1779 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1780 						      priv->port);
1781 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1782 		}
1783 	}
1784 
1785 	/* Detach All multicasts */
1786 	memset(&mc_list[10], 0xff, ETH_ALEN);
1787 	mc_list[5] = priv->port; /* needed for B0 steering support */
1788 	mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1789 			      MLX4_PROT_ETH, priv->broadcast_id);
1790 	list_for_each_entry(mclist, &priv->curr_list, list) {
1791 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1792 		mc_list[5] = priv->port;
1793 		mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1794 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1795 	}
1796 	mlx4_en_clear_list(dev);
1797 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1798 		list_del(&mclist->list);
1799 		kfree(mclist);
1800 	}
1801 
1802 	/* Flush multicast filter */
1803 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1804 
1805 	/* Remove flow steering rules for the port*/
1806 	if (mdev->dev->caps.steering_mode ==
1807 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1808 		ASSERT_RTNL();
1809 		list_for_each_entry_safe(flow, tmp_flow,
1810 					 &priv->ethtool_list, list) {
1811 			mlx4_flow_detach(mdev->dev, flow->id);
1812 			list_del(&flow->list);
1813 		}
1814 	}
1815 
1816 	mlx4_en_destroy_drop_qp(priv);
1817 
1818 	/* Free TX Rings */
1819 	for (i = 0; i < priv->tx_ring_num; i++) {
1820 		mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1821 		mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1822 	}
1823 	msleep(10);
1824 
1825 	for (i = 0; i < priv->tx_ring_num; i++)
1826 		mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1827 
1828 	/* Free RSS qps */
1829 	mlx4_en_release_rss_steer(priv);
1830 
1831 	/* Unregister Mac address for the port */
1832 	mlx4_en_put_qp(priv);
1833 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1834 		mdev->mac_removed[priv->port] = 1;
1835 
1836 	/* Free RX Rings */
1837 	for (i = 0; i < priv->rx_ring_num; i++) {
1838 		struct mlx4_en_cq *cq = priv->rx_cq[i];
1839 
1840 		local_bh_disable();
1841 		while (!mlx4_en_cq_lock_napi(cq)) {
1842 			pr_info("CQ %d locked\n", i);
1843 			mdelay(1);
1844 		}
1845 		local_bh_enable();
1846 
1847 		while (test_bit(NAPI_STATE_SCHED, &cq->napi.state))
1848 			msleep(1);
1849 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1850 		mlx4_en_deactivate_cq(priv, cq);
1851 	}
1852 }
1853 
1854 static void mlx4_en_restart(struct work_struct *work)
1855 {
1856 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1857 						 watchdog_task);
1858 	struct mlx4_en_dev *mdev = priv->mdev;
1859 	struct net_device *dev = priv->dev;
1860 
1861 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1862 
1863 	mutex_lock(&mdev->state_lock);
1864 	if (priv->port_up) {
1865 		mlx4_en_stop_port(dev, 1);
1866 		if (mlx4_en_start_port(dev))
1867 			en_err(priv, "Failed restarting port %d\n", priv->port);
1868 	}
1869 	mutex_unlock(&mdev->state_lock);
1870 }
1871 
1872 static void mlx4_en_clear_stats(struct net_device *dev)
1873 {
1874 	struct mlx4_en_priv *priv = netdev_priv(dev);
1875 	struct mlx4_en_dev *mdev = priv->mdev;
1876 	int i;
1877 
1878 	if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1879 		en_dbg(HW, priv, "Failed dumping statistics\n");
1880 
1881 	memset(&priv->stats, 0, sizeof(priv->stats));
1882 	memset(&priv->pstats, 0, sizeof(priv->pstats));
1883 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1884 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1885 
1886 	for (i = 0; i < priv->tx_ring_num; i++) {
1887 		priv->tx_ring[i]->bytes = 0;
1888 		priv->tx_ring[i]->packets = 0;
1889 		priv->tx_ring[i]->tx_csum = 0;
1890 	}
1891 	for (i = 0; i < priv->rx_ring_num; i++) {
1892 		priv->rx_ring[i]->bytes = 0;
1893 		priv->rx_ring[i]->packets = 0;
1894 		priv->rx_ring[i]->csum_ok = 0;
1895 		priv->rx_ring[i]->csum_none = 0;
1896 	}
1897 }
1898 
1899 static int mlx4_en_open(struct net_device *dev)
1900 {
1901 	struct mlx4_en_priv *priv = netdev_priv(dev);
1902 	struct mlx4_en_dev *mdev = priv->mdev;
1903 	int err = 0;
1904 
1905 	mutex_lock(&mdev->state_lock);
1906 
1907 	if (!mdev->device_up) {
1908 		en_err(priv, "Cannot open - device down/disabled\n");
1909 		err = -EBUSY;
1910 		goto out;
1911 	}
1912 
1913 	/* Reset HW statistics and SW counters */
1914 	mlx4_en_clear_stats(dev);
1915 
1916 	err = mlx4_en_start_port(dev);
1917 	if (err)
1918 		en_err(priv, "Failed starting port:%d\n", priv->port);
1919 
1920 out:
1921 	mutex_unlock(&mdev->state_lock);
1922 	return err;
1923 }
1924 
1925 
1926 static int mlx4_en_close(struct net_device *dev)
1927 {
1928 	struct mlx4_en_priv *priv = netdev_priv(dev);
1929 	struct mlx4_en_dev *mdev = priv->mdev;
1930 
1931 	en_dbg(IFDOWN, priv, "Close port called\n");
1932 
1933 	mutex_lock(&mdev->state_lock);
1934 
1935 	mlx4_en_stop_port(dev, 0);
1936 	netif_carrier_off(dev);
1937 
1938 	mutex_unlock(&mdev->state_lock);
1939 	return 0;
1940 }
1941 
1942 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1943 {
1944 	int i;
1945 
1946 #ifdef CONFIG_RFS_ACCEL
1947 	free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1948 	priv->dev->rx_cpu_rmap = NULL;
1949 #endif
1950 
1951 	for (i = 0; i < priv->tx_ring_num; i++) {
1952 		if (priv->tx_ring && priv->tx_ring[i])
1953 			mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1954 		if (priv->tx_cq && priv->tx_cq[i])
1955 			mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1956 	}
1957 
1958 	for (i = 0; i < priv->rx_ring_num; i++) {
1959 		if (priv->rx_ring[i])
1960 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1961 				priv->prof->rx_ring_size, priv->stride);
1962 		if (priv->rx_cq[i])
1963 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1964 	}
1965 
1966 	if (priv->base_tx_qpn) {
1967 		mlx4_qp_release_range(priv->mdev->dev, priv->base_tx_qpn, priv->tx_ring_num);
1968 		priv->base_tx_qpn = 0;
1969 	}
1970 }
1971 
1972 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1973 {
1974 	struct mlx4_en_port_profile *prof = priv->prof;
1975 	int i;
1976 	int err;
1977 	int node;
1978 
1979 	err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &priv->base_tx_qpn);
1980 	if (err) {
1981 		en_err(priv, "failed reserving range for TX rings\n");
1982 		return err;
1983 	}
1984 
1985 	/* Create tx Rings */
1986 	for (i = 0; i < priv->tx_ring_num; i++) {
1987 		node = cpu_to_node(i % num_online_cpus());
1988 		if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1989 				      prof->tx_ring_size, i, TX, node))
1990 			goto err;
1991 
1992 		if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
1993 					   priv->base_tx_qpn + i,
1994 					   prof->tx_ring_size, TXBB_SIZE,
1995 					   node, i))
1996 			goto err;
1997 	}
1998 
1999 	/* Create rx Rings */
2000 	for (i = 0; i < priv->rx_ring_num; i++) {
2001 		node = cpu_to_node(i % num_online_cpus());
2002 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2003 				      prof->rx_ring_size, i, RX, node))
2004 			goto err;
2005 
2006 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2007 					   prof->rx_ring_size, priv->stride,
2008 					   node))
2009 			goto err;
2010 	}
2011 
2012 #ifdef CONFIG_RFS_ACCEL
2013 	if (priv->mdev->dev->caps.comp_pool) {
2014 		priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->mdev->dev->caps.comp_pool);
2015 		if (!priv->dev->rx_cpu_rmap)
2016 			goto err;
2017 	}
2018 #endif
2019 
2020 	return 0;
2021 
2022 err:
2023 	en_err(priv, "Failed to allocate NIC resources\n");
2024 	for (i = 0; i < priv->rx_ring_num; i++) {
2025 		if (priv->rx_ring[i])
2026 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2027 						prof->rx_ring_size,
2028 						priv->stride);
2029 		if (priv->rx_cq[i])
2030 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2031 	}
2032 	for (i = 0; i < priv->tx_ring_num; i++) {
2033 		if (priv->tx_ring[i])
2034 			mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
2035 		if (priv->tx_cq[i])
2036 			mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
2037 	}
2038 	return -ENOMEM;
2039 }
2040 
2041 
2042 void mlx4_en_destroy_netdev(struct net_device *dev)
2043 {
2044 	struct mlx4_en_priv *priv = netdev_priv(dev);
2045 	struct mlx4_en_dev *mdev = priv->mdev;
2046 
2047 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2048 
2049 	/* Unregister device - this will close the port if it was up */
2050 	if (priv->registered)
2051 		unregister_netdev(dev);
2052 
2053 	if (priv->allocated)
2054 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2055 
2056 	cancel_delayed_work(&priv->stats_task);
2057 	cancel_delayed_work(&priv->service_task);
2058 	/* flush any pending task for this netdev */
2059 	flush_workqueue(mdev->workqueue);
2060 
2061 	/* Detach the netdev so tasks would not attempt to access it */
2062 	mutex_lock(&mdev->state_lock);
2063 	mdev->pndev[priv->port] = NULL;
2064 	mutex_unlock(&mdev->state_lock);
2065 
2066 	mlx4_en_free_resources(priv);
2067 
2068 	kfree(priv->tx_ring);
2069 	kfree(priv->tx_cq);
2070 
2071 	free_netdev(dev);
2072 }
2073 
2074 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2075 {
2076 	struct mlx4_en_priv *priv = netdev_priv(dev);
2077 	struct mlx4_en_dev *mdev = priv->mdev;
2078 	int err = 0;
2079 
2080 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2081 		 dev->mtu, new_mtu);
2082 
2083 	if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
2084 		en_err(priv, "Bad MTU size:%d.\n", new_mtu);
2085 		return -EPERM;
2086 	}
2087 	dev->mtu = new_mtu;
2088 
2089 	if (netif_running(dev)) {
2090 		mutex_lock(&mdev->state_lock);
2091 		if (!mdev->device_up) {
2092 			/* NIC is probably restarting - let watchdog task reset
2093 			 * the port */
2094 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2095 		} else {
2096 			mlx4_en_stop_port(dev, 1);
2097 			err = mlx4_en_start_port(dev);
2098 			if (err) {
2099 				en_err(priv, "Failed restarting port:%d\n",
2100 					 priv->port);
2101 				queue_work(mdev->workqueue, &priv->watchdog_task);
2102 			}
2103 		}
2104 		mutex_unlock(&mdev->state_lock);
2105 	}
2106 	return 0;
2107 }
2108 
2109 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2110 {
2111 	struct mlx4_en_priv *priv = netdev_priv(dev);
2112 	struct mlx4_en_dev *mdev = priv->mdev;
2113 	struct hwtstamp_config config;
2114 
2115 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2116 		return -EFAULT;
2117 
2118 	/* reserved for future extensions */
2119 	if (config.flags)
2120 		return -EINVAL;
2121 
2122 	/* device doesn't support time stamping */
2123 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2124 		return -EINVAL;
2125 
2126 	/* TX HW timestamp */
2127 	switch (config.tx_type) {
2128 	case HWTSTAMP_TX_OFF:
2129 	case HWTSTAMP_TX_ON:
2130 		break;
2131 	default:
2132 		return -ERANGE;
2133 	}
2134 
2135 	/* RX HW timestamp */
2136 	switch (config.rx_filter) {
2137 	case HWTSTAMP_FILTER_NONE:
2138 		break;
2139 	case HWTSTAMP_FILTER_ALL:
2140 	case HWTSTAMP_FILTER_SOME:
2141 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2142 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2143 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2144 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2145 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2146 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2147 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2148 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2149 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2150 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2151 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2152 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2153 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2154 		break;
2155 	default:
2156 		return -ERANGE;
2157 	}
2158 
2159 	if (mlx4_en_timestamp_config(dev, config.tx_type, config.rx_filter)) {
2160 		config.tx_type = HWTSTAMP_TX_OFF;
2161 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2162 	}
2163 
2164 	return copy_to_user(ifr->ifr_data, &config,
2165 			    sizeof(config)) ? -EFAULT : 0;
2166 }
2167 
2168 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2169 {
2170 	struct mlx4_en_priv *priv = netdev_priv(dev);
2171 
2172 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2173 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2174 }
2175 
2176 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2177 {
2178 	switch (cmd) {
2179 	case SIOCSHWTSTAMP:
2180 		return mlx4_en_hwtstamp_set(dev, ifr);
2181 	case SIOCGHWTSTAMP:
2182 		return mlx4_en_hwtstamp_get(dev, ifr);
2183 	default:
2184 		return -EOPNOTSUPP;
2185 	}
2186 }
2187 
2188 static int mlx4_en_set_features(struct net_device *netdev,
2189 		netdev_features_t features)
2190 {
2191 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2192 
2193 	if (features & NETIF_F_LOOPBACK)
2194 		priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
2195 	else
2196 		priv->ctrl_flags &=
2197 			cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
2198 
2199 	mlx4_en_update_loopback_state(netdev, features);
2200 
2201 	return 0;
2202 
2203 }
2204 
2205 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2206 {
2207 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2208 	struct mlx4_en_dev *mdev = en_priv->mdev;
2209 	u64 mac_u64 = mlx4_en_mac_to_u64(mac);
2210 
2211 	if (!is_valid_ether_addr(mac))
2212 		return -EINVAL;
2213 
2214 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2215 }
2216 
2217 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos)
2218 {
2219 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2220 	struct mlx4_en_dev *mdev = en_priv->mdev;
2221 
2222 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos);
2223 }
2224 
2225 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2226 {
2227 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2228 	struct mlx4_en_dev *mdev = en_priv->mdev;
2229 
2230 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2231 }
2232 
2233 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2234 {
2235 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2236 	struct mlx4_en_dev *mdev = en_priv->mdev;
2237 
2238 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2239 }
2240 
2241 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2242 {
2243 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2244 	struct mlx4_en_dev *mdev = en_priv->mdev;
2245 
2246 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2247 }
2248 
2249 #define PORT_ID_BYTE_LEN 8
2250 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2251 				    struct netdev_phys_port_id *ppid)
2252 {
2253 	struct mlx4_en_priv *priv = netdev_priv(dev);
2254 	struct mlx4_dev *mdev = priv->mdev->dev;
2255 	int i;
2256 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2257 
2258 	if (!phys_port_id)
2259 		return -EOPNOTSUPP;
2260 
2261 	ppid->id_len = sizeof(phys_port_id);
2262 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2263 		ppid->id[i] =  phys_port_id & 0xff;
2264 		phys_port_id >>= 8;
2265 	}
2266 	return 0;
2267 }
2268 
2269 static const struct net_device_ops mlx4_netdev_ops = {
2270 	.ndo_open		= mlx4_en_open,
2271 	.ndo_stop		= mlx4_en_close,
2272 	.ndo_start_xmit		= mlx4_en_xmit,
2273 	.ndo_select_queue	= mlx4_en_select_queue,
2274 	.ndo_get_stats		= mlx4_en_get_stats,
2275 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2276 	.ndo_set_mac_address	= mlx4_en_set_mac,
2277 	.ndo_validate_addr	= eth_validate_addr,
2278 	.ndo_change_mtu		= mlx4_en_change_mtu,
2279 	.ndo_do_ioctl		= mlx4_en_ioctl,
2280 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2281 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2282 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2283 #ifdef CONFIG_NET_POLL_CONTROLLER
2284 	.ndo_poll_controller	= mlx4_en_netpoll,
2285 #endif
2286 	.ndo_set_features	= mlx4_en_set_features,
2287 	.ndo_setup_tc		= mlx4_en_setup_tc,
2288 #ifdef CONFIG_RFS_ACCEL
2289 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2290 #endif
2291 #ifdef CONFIG_NET_RX_BUSY_POLL
2292 	.ndo_busy_poll		= mlx4_en_low_latency_recv,
2293 #endif
2294 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2295 };
2296 
2297 static const struct net_device_ops mlx4_netdev_ops_master = {
2298 	.ndo_open		= mlx4_en_open,
2299 	.ndo_stop		= mlx4_en_close,
2300 	.ndo_start_xmit		= mlx4_en_xmit,
2301 	.ndo_select_queue	= mlx4_en_select_queue,
2302 	.ndo_get_stats		= mlx4_en_get_stats,
2303 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2304 	.ndo_set_mac_address	= mlx4_en_set_mac,
2305 	.ndo_validate_addr	= eth_validate_addr,
2306 	.ndo_change_mtu		= mlx4_en_change_mtu,
2307 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2308 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2309 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2310 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2311 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2312 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2313 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2314 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2315 #ifdef CONFIG_NET_POLL_CONTROLLER
2316 	.ndo_poll_controller	= mlx4_en_netpoll,
2317 #endif
2318 	.ndo_set_features	= mlx4_en_set_features,
2319 	.ndo_setup_tc		= mlx4_en_setup_tc,
2320 #ifdef CONFIG_RFS_ACCEL
2321 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2322 #endif
2323 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2324 };
2325 
2326 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2327 			struct mlx4_en_port_profile *prof)
2328 {
2329 	struct net_device *dev;
2330 	struct mlx4_en_priv *priv;
2331 	int i;
2332 	int err;
2333 	u64 mac_u64;
2334 
2335 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2336 				 MAX_TX_RINGS, MAX_RX_RINGS);
2337 	if (dev == NULL)
2338 		return -ENOMEM;
2339 
2340 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2341 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2342 
2343 	SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
2344 	dev->dev_id =  port - 1;
2345 
2346 	/*
2347 	 * Initialize driver private data
2348 	 */
2349 
2350 	priv = netdev_priv(dev);
2351 	memset(priv, 0, sizeof(struct mlx4_en_priv));
2352 	priv->dev = dev;
2353 	priv->mdev = mdev;
2354 	priv->ddev = &mdev->pdev->dev;
2355 	priv->prof = prof;
2356 	priv->port = port;
2357 	priv->port_up = false;
2358 	priv->flags = prof->flags;
2359 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2360 			MLX4_WQE_CTRL_SOLICITED);
2361 	priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2362 	priv->tx_ring_num = prof->tx_ring_num;
2363 
2364 	priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2365 				GFP_KERNEL);
2366 	if (!priv->tx_ring) {
2367 		err = -ENOMEM;
2368 		goto out;
2369 	}
2370 	priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2371 			      GFP_KERNEL);
2372 	if (!priv->tx_cq) {
2373 		err = -ENOMEM;
2374 		goto out;
2375 	}
2376 	priv->rx_ring_num = prof->rx_ring_num;
2377 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2378 	priv->mac_index = -1;
2379 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
2380 	spin_lock_init(&priv->stats_lock);
2381 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2382 	INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2383 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2384 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2385 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2386 #ifdef CONFIG_MLX4_EN_DCB
2387 	if (!mlx4_is_slave(priv->mdev->dev)) {
2388 		if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
2389 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2390 		} else {
2391 			en_info(priv, "enabling only PFC DCB ops\n");
2392 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2393 		}
2394 	}
2395 #endif
2396 
2397 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2398 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
2399 
2400 	/* Query for default mac and max mtu */
2401 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2402 
2403 	/* Set default MAC */
2404 	dev->addr_len = ETH_ALEN;
2405 	mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2406 	if (!is_valid_ether_addr(dev->dev_addr)) {
2407 		if (mlx4_is_slave(priv->mdev->dev)) {
2408 			eth_hw_addr_random(dev);
2409 			en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
2410 			mac_u64 = mlx4_en_mac_to_u64(dev->dev_addr);
2411 			mdev->dev->caps.def_mac[priv->port] = mac_u64;
2412 		} else {
2413 			en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2414 			       priv->port, dev->dev_addr);
2415 			err = -EINVAL;
2416 			goto out;
2417 		}
2418 	}
2419 
2420 	memcpy(priv->prev_mac, dev->dev_addr, sizeof(priv->prev_mac));
2421 
2422 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2423 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2424 	err = mlx4_en_alloc_resources(priv);
2425 	if (err)
2426 		goto out;
2427 
2428 #ifdef CONFIG_RFS_ACCEL
2429 	INIT_LIST_HEAD(&priv->filters);
2430 	spin_lock_init(&priv->filters_lock);
2431 #endif
2432 
2433 	/* Initialize time stamping config */
2434 	priv->hwtstamp_config.flags = 0;
2435 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
2436 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2437 
2438 	/* Allocate page for receive rings */
2439 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2440 				MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2441 	if (err) {
2442 		en_err(priv, "Failed to allocate page for rx qps\n");
2443 		goto out;
2444 	}
2445 	priv->allocated = 1;
2446 
2447 	/*
2448 	 * Initialize netdev entry points
2449 	 */
2450 	if (mlx4_is_master(priv->mdev->dev))
2451 		dev->netdev_ops = &mlx4_netdev_ops_master;
2452 	else
2453 		dev->netdev_ops = &mlx4_netdev_ops;
2454 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2455 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2456 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2457 
2458 	SET_ETHTOOL_OPS(dev, &mlx4_en_ethtool_ops);
2459 
2460 	/*
2461 	 * Set driver features
2462 	 */
2463 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2464 	if (mdev->LSO_support)
2465 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2466 
2467 	dev->vlan_features = dev->hw_features;
2468 
2469 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2470 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2471 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2472 			NETIF_F_HW_VLAN_CTAG_FILTER;
2473 	dev->hw_features |= NETIF_F_LOOPBACK;
2474 
2475 	if (mdev->dev->caps.steering_mode ==
2476 	    MLX4_STEERING_MODE_DEVICE_MANAGED)
2477 		dev->hw_features |= NETIF_F_NTUPLE;
2478 
2479 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
2480 		dev->priv_flags |= IFF_UNICAST_FLT;
2481 
2482 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2483 		dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
2484 					NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL;
2485 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
2486 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL;
2487 	}
2488 
2489 	mdev->pndev[port] = dev;
2490 
2491 	netif_carrier_off(dev);
2492 	mlx4_en_set_default_moderation(priv);
2493 
2494 	err = register_netdev(dev);
2495 	if (err) {
2496 		en_err(priv, "Netdev registration failed for port %d\n", port);
2497 		goto out;
2498 	}
2499 	priv->registered = 1;
2500 
2501 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2502 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2503 
2504 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
2505 
2506 	/* Configure port */
2507 	mlx4_en_calc_rx_buf(dev);
2508 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2509 				    priv->rx_skb_size + ETH_FCS_LEN,
2510 				    prof->tx_pause, prof->tx_ppp,
2511 				    prof->rx_pause, prof->rx_ppp);
2512 	if (err) {
2513 		en_err(priv, "Failed setting port general configurations "
2514 		       "for port %d, with error %d\n", priv->port, err);
2515 		goto out;
2516 	}
2517 
2518 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2519 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC);
2520 		if (err) {
2521 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
2522 			       err);
2523 			goto out;
2524 		}
2525 	}
2526 
2527 	/* Init port */
2528 	en_warn(priv, "Initializing port\n");
2529 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
2530 	if (err) {
2531 		en_err(priv, "Failed Initializing port\n");
2532 		goto out;
2533 	}
2534 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2535 
2536 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2537 		queue_delayed_work(mdev->workqueue, &priv->service_task,
2538 				   SERVICE_TASK_DELAY);
2539 
2540 	return 0;
2541 
2542 out:
2543 	mlx4_en_destroy_netdev(dev);
2544 	return err;
2545 }
2546 
2547