xref: /linux/drivers/net/ethernet/mellanox/mlx4/en_netdev.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 #include <net/vxlan.h>
43 
44 #include <linux/mlx4/driver.h>
45 #include <linux/mlx4/device.h>
46 #include <linux/mlx4/cmd.h>
47 #include <linux/mlx4/cq.h>
48 
49 #include "mlx4_en.h"
50 #include "en_port.h"
51 
52 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
53 {
54 	struct mlx4_en_priv *priv = netdev_priv(dev);
55 	int i;
56 	unsigned int offset = 0;
57 
58 	if (up && up != MLX4_EN_NUM_UP)
59 		return -EINVAL;
60 
61 	netdev_set_num_tc(dev, up);
62 
63 	/* Partition Tx queues evenly amongst UP's */
64 	for (i = 0; i < up; i++) {
65 		netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
66 		offset += priv->num_tx_rings_p_up;
67 	}
68 
69 	return 0;
70 }
71 
72 #ifdef CONFIG_NET_RX_BUSY_POLL
73 /* must be called with local_bh_disable()d */
74 static int mlx4_en_low_latency_recv(struct napi_struct *napi)
75 {
76 	struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
77 	struct net_device *dev = cq->dev;
78 	struct mlx4_en_priv *priv = netdev_priv(dev);
79 	struct mlx4_en_rx_ring *rx_ring = priv->rx_ring[cq->ring];
80 	int done;
81 
82 	if (!priv->port_up)
83 		return LL_FLUSH_FAILED;
84 
85 	if (!mlx4_en_cq_lock_poll(cq))
86 		return LL_FLUSH_BUSY;
87 
88 	done = mlx4_en_process_rx_cq(dev, cq, 4);
89 	if (likely(done))
90 		rx_ring->cleaned += done;
91 	else
92 		rx_ring->misses++;
93 
94 	mlx4_en_cq_unlock_poll(cq);
95 
96 	return done;
97 }
98 #endif	/* CONFIG_NET_RX_BUSY_POLL */
99 
100 #ifdef CONFIG_RFS_ACCEL
101 
102 struct mlx4_en_filter {
103 	struct list_head next;
104 	struct work_struct work;
105 
106 	u8     ip_proto;
107 	__be32 src_ip;
108 	__be32 dst_ip;
109 	__be16 src_port;
110 	__be16 dst_port;
111 
112 	int rxq_index;
113 	struct mlx4_en_priv *priv;
114 	u32 flow_id;			/* RFS infrastructure id */
115 	int id;				/* mlx4_en driver id */
116 	u64 reg_id;			/* Flow steering API id */
117 	u8 activated;			/* Used to prevent expiry before filter
118 					 * is attached
119 					 */
120 	struct hlist_node filter_chain;
121 };
122 
123 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
124 
125 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
126 {
127 	switch (ip_proto) {
128 	case IPPROTO_UDP:
129 		return MLX4_NET_TRANS_RULE_ID_UDP;
130 	case IPPROTO_TCP:
131 		return MLX4_NET_TRANS_RULE_ID_TCP;
132 	default:
133 		return MLX4_NET_TRANS_RULE_NUM;
134 	}
135 };
136 
137 static void mlx4_en_filter_work(struct work_struct *work)
138 {
139 	struct mlx4_en_filter *filter = container_of(work,
140 						     struct mlx4_en_filter,
141 						     work);
142 	struct mlx4_en_priv *priv = filter->priv;
143 	struct mlx4_spec_list spec_tcp_udp = {
144 		.id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
145 		{
146 			.tcp_udp = {
147 				.dst_port = filter->dst_port,
148 				.dst_port_msk = (__force __be16)-1,
149 				.src_port = filter->src_port,
150 				.src_port_msk = (__force __be16)-1,
151 			},
152 		},
153 	};
154 	struct mlx4_spec_list spec_ip = {
155 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
156 		{
157 			.ipv4 = {
158 				.dst_ip = filter->dst_ip,
159 				.dst_ip_msk = (__force __be32)-1,
160 				.src_ip = filter->src_ip,
161 				.src_ip_msk = (__force __be32)-1,
162 			},
163 		},
164 	};
165 	struct mlx4_spec_list spec_eth = {
166 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
167 	};
168 	struct mlx4_net_trans_rule rule = {
169 		.list = LIST_HEAD_INIT(rule.list),
170 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
171 		.exclusive = 1,
172 		.allow_loopback = 1,
173 		.promisc_mode = MLX4_FS_REGULAR,
174 		.port = priv->port,
175 		.priority = MLX4_DOMAIN_RFS,
176 	};
177 	int rc;
178 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
179 
180 	if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
181 		en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
182 			filter->ip_proto);
183 		goto ignore;
184 	}
185 	list_add_tail(&spec_eth.list, &rule.list);
186 	list_add_tail(&spec_ip.list, &rule.list);
187 	list_add_tail(&spec_tcp_udp.list, &rule.list);
188 
189 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
190 	memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
191 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
192 
193 	filter->activated = 0;
194 
195 	if (filter->reg_id) {
196 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
197 		if (rc && rc != -ENOENT)
198 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
199 	}
200 
201 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
202 	if (rc)
203 		en_err(priv, "Error attaching flow. err = %d\n", rc);
204 
205 ignore:
206 	mlx4_en_filter_rfs_expire(priv);
207 
208 	filter->activated = 1;
209 }
210 
211 static inline struct hlist_head *
212 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
213 		   __be16 src_port, __be16 dst_port)
214 {
215 	unsigned long l;
216 	int bucket_idx;
217 
218 	l = (__force unsigned long)src_port |
219 	    ((__force unsigned long)dst_port << 2);
220 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
221 
222 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
223 
224 	return &priv->filter_hash[bucket_idx];
225 }
226 
227 static struct mlx4_en_filter *
228 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
229 		     __be32 dst_ip, u8 ip_proto, __be16 src_port,
230 		     __be16 dst_port, u32 flow_id)
231 {
232 	struct mlx4_en_filter *filter = NULL;
233 
234 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
235 	if (!filter)
236 		return NULL;
237 
238 	filter->priv = priv;
239 	filter->rxq_index = rxq_index;
240 	INIT_WORK(&filter->work, mlx4_en_filter_work);
241 
242 	filter->src_ip = src_ip;
243 	filter->dst_ip = dst_ip;
244 	filter->ip_proto = ip_proto;
245 	filter->src_port = src_port;
246 	filter->dst_port = dst_port;
247 
248 	filter->flow_id = flow_id;
249 
250 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
251 
252 	list_add_tail(&filter->next, &priv->filters);
253 	hlist_add_head(&filter->filter_chain,
254 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
255 					  dst_port));
256 
257 	return filter;
258 }
259 
260 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
261 {
262 	struct mlx4_en_priv *priv = filter->priv;
263 	int rc;
264 
265 	list_del(&filter->next);
266 
267 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
268 	if (rc && rc != -ENOENT)
269 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
270 
271 	kfree(filter);
272 }
273 
274 static inline struct mlx4_en_filter *
275 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
276 		    u8 ip_proto, __be16 src_port, __be16 dst_port)
277 {
278 	struct mlx4_en_filter *filter;
279 	struct mlx4_en_filter *ret = NULL;
280 
281 	hlist_for_each_entry(filter,
282 			     filter_hash_bucket(priv, src_ip, dst_ip,
283 						src_port, dst_port),
284 			     filter_chain) {
285 		if (filter->src_ip == src_ip &&
286 		    filter->dst_ip == dst_ip &&
287 		    filter->ip_proto == ip_proto &&
288 		    filter->src_port == src_port &&
289 		    filter->dst_port == dst_port) {
290 			ret = filter;
291 			break;
292 		}
293 	}
294 
295 	return ret;
296 }
297 
298 static int
299 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
300 		   u16 rxq_index, u32 flow_id)
301 {
302 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
303 	struct mlx4_en_filter *filter;
304 	const struct iphdr *ip;
305 	const __be16 *ports;
306 	u8 ip_proto;
307 	__be32 src_ip;
308 	__be32 dst_ip;
309 	__be16 src_port;
310 	__be16 dst_port;
311 	int nhoff = skb_network_offset(skb);
312 	int ret = 0;
313 
314 	if (skb->protocol != htons(ETH_P_IP))
315 		return -EPROTONOSUPPORT;
316 
317 	ip = (const struct iphdr *)(skb->data + nhoff);
318 	if (ip_is_fragment(ip))
319 		return -EPROTONOSUPPORT;
320 
321 	if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
322 		return -EPROTONOSUPPORT;
323 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
324 
325 	ip_proto = ip->protocol;
326 	src_ip = ip->saddr;
327 	dst_ip = ip->daddr;
328 	src_port = ports[0];
329 	dst_port = ports[1];
330 
331 	spin_lock_bh(&priv->filters_lock);
332 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
333 				     src_port, dst_port);
334 	if (filter) {
335 		if (filter->rxq_index == rxq_index)
336 			goto out;
337 
338 		filter->rxq_index = rxq_index;
339 	} else {
340 		filter = mlx4_en_filter_alloc(priv, rxq_index,
341 					      src_ip, dst_ip, ip_proto,
342 					      src_port, dst_port, flow_id);
343 		if (!filter) {
344 			ret = -ENOMEM;
345 			goto err;
346 		}
347 	}
348 
349 	queue_work(priv->mdev->workqueue, &filter->work);
350 
351 out:
352 	ret = filter->id;
353 err:
354 	spin_unlock_bh(&priv->filters_lock);
355 
356 	return ret;
357 }
358 
359 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
360 {
361 	struct mlx4_en_filter *filter, *tmp;
362 	LIST_HEAD(del_list);
363 
364 	spin_lock_bh(&priv->filters_lock);
365 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
366 		list_move(&filter->next, &del_list);
367 		hlist_del(&filter->filter_chain);
368 	}
369 	spin_unlock_bh(&priv->filters_lock);
370 
371 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
372 		cancel_work_sync(&filter->work);
373 		mlx4_en_filter_free(filter);
374 	}
375 }
376 
377 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
378 {
379 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
380 	LIST_HEAD(del_list);
381 	int i = 0;
382 
383 	spin_lock_bh(&priv->filters_lock);
384 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
385 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
386 			break;
387 
388 		if (filter->activated &&
389 		    !work_pending(&filter->work) &&
390 		    rps_may_expire_flow(priv->dev,
391 					filter->rxq_index, filter->flow_id,
392 					filter->id)) {
393 			list_move(&filter->next, &del_list);
394 			hlist_del(&filter->filter_chain);
395 		} else
396 			last_filter = filter;
397 
398 		i++;
399 	}
400 
401 	if (last_filter && (&last_filter->next != priv->filters.next))
402 		list_move(&priv->filters, &last_filter->next);
403 
404 	spin_unlock_bh(&priv->filters_lock);
405 
406 	list_for_each_entry_safe(filter, tmp, &del_list, next)
407 		mlx4_en_filter_free(filter);
408 }
409 #endif
410 
411 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
412 				   __be16 proto, u16 vid)
413 {
414 	struct mlx4_en_priv *priv = netdev_priv(dev);
415 	struct mlx4_en_dev *mdev = priv->mdev;
416 	int err;
417 	int idx;
418 
419 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
420 
421 	set_bit(vid, priv->active_vlans);
422 
423 	/* Add VID to port VLAN filter */
424 	mutex_lock(&mdev->state_lock);
425 	if (mdev->device_up && priv->port_up) {
426 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
427 		if (err)
428 			en_err(priv, "Failed configuring VLAN filter\n");
429 	}
430 	if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
431 		en_dbg(HW, priv, "failed adding vlan %d\n", vid);
432 	mutex_unlock(&mdev->state_lock);
433 
434 	return 0;
435 }
436 
437 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
438 				    __be16 proto, u16 vid)
439 {
440 	struct mlx4_en_priv *priv = netdev_priv(dev);
441 	struct mlx4_en_dev *mdev = priv->mdev;
442 	int err;
443 
444 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
445 
446 	clear_bit(vid, priv->active_vlans);
447 
448 	/* Remove VID from port VLAN filter */
449 	mutex_lock(&mdev->state_lock);
450 	mlx4_unregister_vlan(mdev->dev, priv->port, vid);
451 
452 	if (mdev->device_up && priv->port_up) {
453 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
454 		if (err)
455 			en_err(priv, "Failed configuring VLAN filter\n");
456 	}
457 	mutex_unlock(&mdev->state_lock);
458 
459 	return 0;
460 }
461 
462 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
463 {
464 	int i;
465 	for (i = ETH_ALEN - 1; i >= 0; --i) {
466 		dst_mac[i] = src_mac & 0xff;
467 		src_mac >>= 8;
468 	}
469 	memset(&dst_mac[ETH_ALEN], 0, 2);
470 }
471 
472 
473 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
474 				    int qpn, u64 *reg_id)
475 {
476 	int err;
477 
478 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
479 	    priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
480 		return 0; /* do nothing */
481 
482 	err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
483 				    MLX4_DOMAIN_NIC, reg_id);
484 	if (err) {
485 		en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
486 		return err;
487 	}
488 	en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
489 	return 0;
490 }
491 
492 
493 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
494 				unsigned char *mac, int *qpn, u64 *reg_id)
495 {
496 	struct mlx4_en_dev *mdev = priv->mdev;
497 	struct mlx4_dev *dev = mdev->dev;
498 	int err;
499 
500 	switch (dev->caps.steering_mode) {
501 	case MLX4_STEERING_MODE_B0: {
502 		struct mlx4_qp qp;
503 		u8 gid[16] = {0};
504 
505 		qp.qpn = *qpn;
506 		memcpy(&gid[10], mac, ETH_ALEN);
507 		gid[5] = priv->port;
508 
509 		err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
510 		break;
511 	}
512 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
513 		struct mlx4_spec_list spec_eth = { {NULL} };
514 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
515 
516 		struct mlx4_net_trans_rule rule = {
517 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
518 			.exclusive = 0,
519 			.allow_loopback = 1,
520 			.promisc_mode = MLX4_FS_REGULAR,
521 			.priority = MLX4_DOMAIN_NIC,
522 		};
523 
524 		rule.port = priv->port;
525 		rule.qpn = *qpn;
526 		INIT_LIST_HEAD(&rule.list);
527 
528 		spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
529 		memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
530 		memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
531 		list_add_tail(&spec_eth.list, &rule.list);
532 
533 		err = mlx4_flow_attach(dev, &rule, reg_id);
534 		break;
535 	}
536 	default:
537 		return -EINVAL;
538 	}
539 	if (err)
540 		en_warn(priv, "Failed Attaching Unicast\n");
541 
542 	return err;
543 }
544 
545 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
546 				     unsigned char *mac, int qpn, u64 reg_id)
547 {
548 	struct mlx4_en_dev *mdev = priv->mdev;
549 	struct mlx4_dev *dev = mdev->dev;
550 
551 	switch (dev->caps.steering_mode) {
552 	case MLX4_STEERING_MODE_B0: {
553 		struct mlx4_qp qp;
554 		u8 gid[16] = {0};
555 
556 		qp.qpn = qpn;
557 		memcpy(&gid[10], mac, ETH_ALEN);
558 		gid[5] = priv->port;
559 
560 		mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
561 		break;
562 	}
563 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
564 		mlx4_flow_detach(dev, reg_id);
565 		break;
566 	}
567 	default:
568 		en_err(priv, "Invalid steering mode.\n");
569 	}
570 }
571 
572 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
573 {
574 	struct mlx4_en_dev *mdev = priv->mdev;
575 	struct mlx4_dev *dev = mdev->dev;
576 	struct mlx4_mac_entry *entry;
577 	int index = 0;
578 	int err = 0;
579 	u64 reg_id = 0;
580 	int *qpn = &priv->base_qpn;
581 	u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
582 
583 	en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
584 	       priv->dev->dev_addr);
585 	index = mlx4_register_mac(dev, priv->port, mac);
586 	if (index < 0) {
587 		err = index;
588 		en_err(priv, "Failed adding MAC: %pM\n",
589 		       priv->dev->dev_addr);
590 		return err;
591 	}
592 
593 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
594 		int base_qpn = mlx4_get_base_qpn(dev, priv->port);
595 		*qpn = base_qpn + index;
596 		return 0;
597 	}
598 
599 	err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP);
600 	en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
601 	if (err) {
602 		en_err(priv, "Failed to reserve qp for mac registration\n");
603 		goto qp_err;
604 	}
605 
606 	err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
607 	if (err)
608 		goto steer_err;
609 
610 	err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
611 				       &priv->tunnel_reg_id);
612 	if (err)
613 		goto tunnel_err;
614 
615 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
616 	if (!entry) {
617 		err = -ENOMEM;
618 		goto alloc_err;
619 	}
620 	memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
621 	memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
622 	entry->reg_id = reg_id;
623 
624 	hlist_add_head_rcu(&entry->hlist,
625 			   &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
626 
627 	return 0;
628 
629 alloc_err:
630 	if (priv->tunnel_reg_id)
631 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
632 tunnel_err:
633 	mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
634 
635 steer_err:
636 	mlx4_qp_release_range(dev, *qpn, 1);
637 
638 qp_err:
639 	mlx4_unregister_mac(dev, priv->port, mac);
640 	return err;
641 }
642 
643 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
644 {
645 	struct mlx4_en_dev *mdev = priv->mdev;
646 	struct mlx4_dev *dev = mdev->dev;
647 	int qpn = priv->base_qpn;
648 	u64 mac;
649 
650 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
651 		mac = mlx4_mac_to_u64(priv->dev->dev_addr);
652 		en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
653 		       priv->dev->dev_addr);
654 		mlx4_unregister_mac(dev, priv->port, mac);
655 	} else {
656 		struct mlx4_mac_entry *entry;
657 		struct hlist_node *tmp;
658 		struct hlist_head *bucket;
659 		unsigned int i;
660 
661 		for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
662 			bucket = &priv->mac_hash[i];
663 			hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
664 				mac = mlx4_mac_to_u64(entry->mac);
665 				en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
666 				       entry->mac);
667 				mlx4_en_uc_steer_release(priv, entry->mac,
668 							 qpn, entry->reg_id);
669 
670 				mlx4_unregister_mac(dev, priv->port, mac);
671 				hlist_del_rcu(&entry->hlist);
672 				kfree_rcu(entry, rcu);
673 			}
674 		}
675 
676 		if (priv->tunnel_reg_id) {
677 			mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
678 			priv->tunnel_reg_id = 0;
679 		}
680 
681 		en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
682 		       priv->port, qpn);
683 		mlx4_qp_release_range(dev, qpn, 1);
684 		priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
685 	}
686 }
687 
688 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
689 			       unsigned char *new_mac, unsigned char *prev_mac)
690 {
691 	struct mlx4_en_dev *mdev = priv->mdev;
692 	struct mlx4_dev *dev = mdev->dev;
693 	int err = 0;
694 	u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
695 
696 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
697 		struct hlist_head *bucket;
698 		unsigned int mac_hash;
699 		struct mlx4_mac_entry *entry;
700 		struct hlist_node *tmp;
701 		u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
702 
703 		bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
704 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
705 			if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
706 				mlx4_en_uc_steer_release(priv, entry->mac,
707 							 qpn, entry->reg_id);
708 				mlx4_unregister_mac(dev, priv->port,
709 						    prev_mac_u64);
710 				hlist_del_rcu(&entry->hlist);
711 				synchronize_rcu();
712 				memcpy(entry->mac, new_mac, ETH_ALEN);
713 				entry->reg_id = 0;
714 				mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
715 				hlist_add_head_rcu(&entry->hlist,
716 						   &priv->mac_hash[mac_hash]);
717 				mlx4_register_mac(dev, priv->port, new_mac_u64);
718 				err = mlx4_en_uc_steer_add(priv, new_mac,
719 							   &qpn,
720 							   &entry->reg_id);
721 				if (err)
722 					return err;
723 				if (priv->tunnel_reg_id) {
724 					mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
725 					priv->tunnel_reg_id = 0;
726 				}
727 				err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
728 							       &priv->tunnel_reg_id);
729 				return err;
730 			}
731 		}
732 		return -EINVAL;
733 	}
734 
735 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
736 }
737 
738 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
739 			      unsigned char new_mac[ETH_ALEN + 2])
740 {
741 	int err = 0;
742 
743 	if (priv->port_up) {
744 		/* Remove old MAC and insert the new one */
745 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
746 					  new_mac, priv->current_mac);
747 		if (err)
748 			en_err(priv, "Failed changing HW MAC address\n");
749 	} else
750 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
751 
752 	if (!err)
753 		memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
754 
755 	return err;
756 }
757 
758 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
759 {
760 	struct mlx4_en_priv *priv = netdev_priv(dev);
761 	struct mlx4_en_dev *mdev = priv->mdev;
762 	struct sockaddr *saddr = addr;
763 	unsigned char new_mac[ETH_ALEN + 2];
764 	int err;
765 
766 	if (!is_valid_ether_addr(saddr->sa_data))
767 		return -EADDRNOTAVAIL;
768 
769 	mutex_lock(&mdev->state_lock);
770 	memcpy(new_mac, saddr->sa_data, ETH_ALEN);
771 	err = mlx4_en_do_set_mac(priv, new_mac);
772 	if (!err)
773 		memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
774 	mutex_unlock(&mdev->state_lock);
775 
776 	return err;
777 }
778 
779 static void mlx4_en_clear_list(struct net_device *dev)
780 {
781 	struct mlx4_en_priv *priv = netdev_priv(dev);
782 	struct mlx4_en_mc_list *tmp, *mc_to_del;
783 
784 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
785 		list_del(&mc_to_del->list);
786 		kfree(mc_to_del);
787 	}
788 }
789 
790 static void mlx4_en_cache_mclist(struct net_device *dev)
791 {
792 	struct mlx4_en_priv *priv = netdev_priv(dev);
793 	struct netdev_hw_addr *ha;
794 	struct mlx4_en_mc_list *tmp;
795 
796 	mlx4_en_clear_list(dev);
797 	netdev_for_each_mc_addr(ha, dev) {
798 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
799 		if (!tmp) {
800 			mlx4_en_clear_list(dev);
801 			return;
802 		}
803 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
804 		list_add_tail(&tmp->list, &priv->mc_list);
805 	}
806 }
807 
808 static void update_mclist_flags(struct mlx4_en_priv *priv,
809 				struct list_head *dst,
810 				struct list_head *src)
811 {
812 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
813 	bool found;
814 
815 	/* Find all the entries that should be removed from dst,
816 	 * These are the entries that are not found in src
817 	 */
818 	list_for_each_entry(dst_tmp, dst, list) {
819 		found = false;
820 		list_for_each_entry(src_tmp, src, list) {
821 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
822 				found = true;
823 				break;
824 			}
825 		}
826 		if (!found)
827 			dst_tmp->action = MCLIST_REM;
828 	}
829 
830 	/* Add entries that exist in src but not in dst
831 	 * mark them as need to add
832 	 */
833 	list_for_each_entry(src_tmp, src, list) {
834 		found = false;
835 		list_for_each_entry(dst_tmp, dst, list) {
836 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
837 				dst_tmp->action = MCLIST_NONE;
838 				found = true;
839 				break;
840 			}
841 		}
842 		if (!found) {
843 			new_mc = kmemdup(src_tmp,
844 					 sizeof(struct mlx4_en_mc_list),
845 					 GFP_KERNEL);
846 			if (!new_mc)
847 				return;
848 
849 			new_mc->action = MCLIST_ADD;
850 			list_add_tail(&new_mc->list, dst);
851 		}
852 	}
853 }
854 
855 static void mlx4_en_set_rx_mode(struct net_device *dev)
856 {
857 	struct mlx4_en_priv *priv = netdev_priv(dev);
858 
859 	if (!priv->port_up)
860 		return;
861 
862 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
863 }
864 
865 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
866 				     struct mlx4_en_dev *mdev)
867 {
868 	int err = 0;
869 
870 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
871 		if (netif_msg_rx_status(priv))
872 			en_warn(priv, "Entering promiscuous mode\n");
873 		priv->flags |= MLX4_EN_FLAG_PROMISC;
874 
875 		/* Enable promiscouos mode */
876 		switch (mdev->dev->caps.steering_mode) {
877 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
878 			err = mlx4_flow_steer_promisc_add(mdev->dev,
879 							  priv->port,
880 							  priv->base_qpn,
881 							  MLX4_FS_ALL_DEFAULT);
882 			if (err)
883 				en_err(priv, "Failed enabling promiscuous mode\n");
884 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
885 			break;
886 
887 		case MLX4_STEERING_MODE_B0:
888 			err = mlx4_unicast_promisc_add(mdev->dev,
889 						       priv->base_qpn,
890 						       priv->port);
891 			if (err)
892 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
893 
894 			/* Add the default qp number as multicast
895 			 * promisc
896 			 */
897 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
898 				err = mlx4_multicast_promisc_add(mdev->dev,
899 								 priv->base_qpn,
900 								 priv->port);
901 				if (err)
902 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
903 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
904 			}
905 			break;
906 
907 		case MLX4_STEERING_MODE_A0:
908 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
909 						     priv->port,
910 						     priv->base_qpn,
911 						     1);
912 			if (err)
913 				en_err(priv, "Failed enabling promiscuous mode\n");
914 			break;
915 		}
916 
917 		/* Disable port multicast filter (unconditionally) */
918 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
919 					  0, MLX4_MCAST_DISABLE);
920 		if (err)
921 			en_err(priv, "Failed disabling multicast filter\n");
922 	}
923 }
924 
925 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
926 				       struct mlx4_en_dev *mdev)
927 {
928 	int err = 0;
929 
930 	if (netif_msg_rx_status(priv))
931 		en_warn(priv, "Leaving promiscuous mode\n");
932 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
933 
934 	/* Disable promiscouos mode */
935 	switch (mdev->dev->caps.steering_mode) {
936 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
937 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
938 						     priv->port,
939 						     MLX4_FS_ALL_DEFAULT);
940 		if (err)
941 			en_err(priv, "Failed disabling promiscuous mode\n");
942 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
943 		break;
944 
945 	case MLX4_STEERING_MODE_B0:
946 		err = mlx4_unicast_promisc_remove(mdev->dev,
947 						  priv->base_qpn,
948 						  priv->port);
949 		if (err)
950 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
951 		/* Disable Multicast promisc */
952 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
953 			err = mlx4_multicast_promisc_remove(mdev->dev,
954 							    priv->base_qpn,
955 							    priv->port);
956 			if (err)
957 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
958 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
959 		}
960 		break;
961 
962 	case MLX4_STEERING_MODE_A0:
963 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
964 					     priv->port,
965 					     priv->base_qpn, 0);
966 		if (err)
967 			en_err(priv, "Failed disabling promiscuous mode\n");
968 		break;
969 	}
970 }
971 
972 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
973 				 struct net_device *dev,
974 				 struct mlx4_en_dev *mdev)
975 {
976 	struct mlx4_en_mc_list *mclist, *tmp;
977 	u64 mcast_addr = 0;
978 	u8 mc_list[16] = {0};
979 	int err = 0;
980 
981 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
982 	if (dev->flags & IFF_ALLMULTI) {
983 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
984 					  0, MLX4_MCAST_DISABLE);
985 		if (err)
986 			en_err(priv, "Failed disabling multicast filter\n");
987 
988 		/* Add the default qp number as multicast promisc */
989 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
990 			switch (mdev->dev->caps.steering_mode) {
991 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
992 				err = mlx4_flow_steer_promisc_add(mdev->dev,
993 								  priv->port,
994 								  priv->base_qpn,
995 								  MLX4_FS_MC_DEFAULT);
996 				break;
997 
998 			case MLX4_STEERING_MODE_B0:
999 				err = mlx4_multicast_promisc_add(mdev->dev,
1000 								 priv->base_qpn,
1001 								 priv->port);
1002 				break;
1003 
1004 			case MLX4_STEERING_MODE_A0:
1005 				break;
1006 			}
1007 			if (err)
1008 				en_err(priv, "Failed entering multicast promisc mode\n");
1009 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1010 		}
1011 	} else {
1012 		/* Disable Multicast promisc */
1013 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1014 			switch (mdev->dev->caps.steering_mode) {
1015 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1016 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
1017 								     priv->port,
1018 								     MLX4_FS_MC_DEFAULT);
1019 				break;
1020 
1021 			case MLX4_STEERING_MODE_B0:
1022 				err = mlx4_multicast_promisc_remove(mdev->dev,
1023 								    priv->base_qpn,
1024 								    priv->port);
1025 				break;
1026 
1027 			case MLX4_STEERING_MODE_A0:
1028 				break;
1029 			}
1030 			if (err)
1031 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
1032 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1033 		}
1034 
1035 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1036 					  0, MLX4_MCAST_DISABLE);
1037 		if (err)
1038 			en_err(priv, "Failed disabling multicast filter\n");
1039 
1040 		/* Flush mcast filter and init it with broadcast address */
1041 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1042 				    1, MLX4_MCAST_CONFIG);
1043 
1044 		/* Update multicast list - we cache all addresses so they won't
1045 		 * change while HW is updated holding the command semaphor */
1046 		netif_addr_lock_bh(dev);
1047 		mlx4_en_cache_mclist(dev);
1048 		netif_addr_unlock_bh(dev);
1049 		list_for_each_entry(mclist, &priv->mc_list, list) {
1050 			mcast_addr = mlx4_mac_to_u64(mclist->addr);
1051 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1052 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
1053 		}
1054 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1055 					  0, MLX4_MCAST_ENABLE);
1056 		if (err)
1057 			en_err(priv, "Failed enabling multicast filter\n");
1058 
1059 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1060 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1061 			if (mclist->action == MCLIST_REM) {
1062 				/* detach this address and delete from list */
1063 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1064 				mc_list[5] = priv->port;
1065 				err = mlx4_multicast_detach(mdev->dev,
1066 							    &priv->rss_map.indir_qp,
1067 							    mc_list,
1068 							    MLX4_PROT_ETH,
1069 							    mclist->reg_id);
1070 				if (err)
1071 					en_err(priv, "Fail to detach multicast address\n");
1072 
1073 				if (mclist->tunnel_reg_id) {
1074 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1075 					if (err)
1076 						en_err(priv, "Failed to detach multicast address\n");
1077 				}
1078 
1079 				/* remove from list */
1080 				list_del(&mclist->list);
1081 				kfree(mclist);
1082 			} else if (mclist->action == MCLIST_ADD) {
1083 				/* attach the address */
1084 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1085 				/* needed for B0 steering support */
1086 				mc_list[5] = priv->port;
1087 				err = mlx4_multicast_attach(mdev->dev,
1088 							    &priv->rss_map.indir_qp,
1089 							    mc_list,
1090 							    priv->port, 0,
1091 							    MLX4_PROT_ETH,
1092 							    &mclist->reg_id);
1093 				if (err)
1094 					en_err(priv, "Fail to attach multicast address\n");
1095 
1096 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1097 							       &mclist->tunnel_reg_id);
1098 				if (err)
1099 					en_err(priv, "Failed to attach multicast address\n");
1100 			}
1101 		}
1102 	}
1103 }
1104 
1105 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1106 				 struct net_device *dev,
1107 				 struct mlx4_en_dev *mdev)
1108 {
1109 	struct netdev_hw_addr *ha;
1110 	struct mlx4_mac_entry *entry;
1111 	struct hlist_node *tmp;
1112 	bool found;
1113 	u64 mac;
1114 	int err = 0;
1115 	struct hlist_head *bucket;
1116 	unsigned int i;
1117 	int removed = 0;
1118 	u32 prev_flags;
1119 
1120 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1121 	 * since all modification code is protected by mdev->state_lock
1122 	 */
1123 
1124 	/* find what to remove */
1125 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1126 		bucket = &priv->mac_hash[i];
1127 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1128 			found = false;
1129 			netdev_for_each_uc_addr(ha, dev) {
1130 				if (ether_addr_equal_64bits(entry->mac,
1131 							    ha->addr)) {
1132 					found = true;
1133 					break;
1134 				}
1135 			}
1136 
1137 			/* MAC address of the port is not in uc list */
1138 			if (ether_addr_equal_64bits(entry->mac,
1139 						    priv->current_mac))
1140 				found = true;
1141 
1142 			if (!found) {
1143 				mac = mlx4_mac_to_u64(entry->mac);
1144 				mlx4_en_uc_steer_release(priv, entry->mac,
1145 							 priv->base_qpn,
1146 							 entry->reg_id);
1147 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1148 
1149 				hlist_del_rcu(&entry->hlist);
1150 				kfree_rcu(entry, rcu);
1151 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1152 				       entry->mac, priv->port);
1153 				++removed;
1154 			}
1155 		}
1156 	}
1157 
1158 	/* if we didn't remove anything, there is no use in trying to add
1159 	 * again once we are in a forced promisc mode state
1160 	 */
1161 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1162 		return;
1163 
1164 	prev_flags = priv->flags;
1165 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1166 
1167 	/* find what to add */
1168 	netdev_for_each_uc_addr(ha, dev) {
1169 		found = false;
1170 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1171 		hlist_for_each_entry(entry, bucket, hlist) {
1172 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1173 				found = true;
1174 				break;
1175 			}
1176 		}
1177 
1178 		if (!found) {
1179 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1180 			if (!entry) {
1181 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1182 				       ha->addr, priv->port);
1183 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1184 				break;
1185 			}
1186 			mac = mlx4_mac_to_u64(ha->addr);
1187 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1188 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1189 			if (err < 0) {
1190 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1191 				       ha->addr, priv->port, err);
1192 				kfree(entry);
1193 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1194 				break;
1195 			}
1196 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1197 						   &priv->base_qpn,
1198 						   &entry->reg_id);
1199 			if (err) {
1200 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1201 				       ha->addr, priv->port, err);
1202 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1203 				kfree(entry);
1204 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1205 				break;
1206 			} else {
1207 				unsigned int mac_hash;
1208 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1209 				       ha->addr, priv->port);
1210 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1211 				bucket = &priv->mac_hash[mac_hash];
1212 				hlist_add_head_rcu(&entry->hlist, bucket);
1213 			}
1214 		}
1215 	}
1216 
1217 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1218 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1219 			priv->port);
1220 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1221 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1222 			priv->port);
1223 	}
1224 }
1225 
1226 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1227 {
1228 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1229 						 rx_mode_task);
1230 	struct mlx4_en_dev *mdev = priv->mdev;
1231 	struct net_device *dev = priv->dev;
1232 
1233 	mutex_lock(&mdev->state_lock);
1234 	if (!mdev->device_up) {
1235 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1236 		goto out;
1237 	}
1238 	if (!priv->port_up) {
1239 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1240 		goto out;
1241 	}
1242 
1243 	if (!netif_carrier_ok(dev)) {
1244 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1245 			if (priv->port_state.link_state) {
1246 				priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1247 				netif_carrier_on(dev);
1248 				en_dbg(LINK, priv, "Link Up\n");
1249 			}
1250 		}
1251 	}
1252 
1253 	if (dev->priv_flags & IFF_UNICAST_FLT)
1254 		mlx4_en_do_uc_filter(priv, dev, mdev);
1255 
1256 	/* Promsicuous mode: disable all filters */
1257 	if ((dev->flags & IFF_PROMISC) ||
1258 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1259 		mlx4_en_set_promisc_mode(priv, mdev);
1260 		goto out;
1261 	}
1262 
1263 	/* Not in promiscuous mode */
1264 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1265 		mlx4_en_clear_promisc_mode(priv, mdev);
1266 
1267 	mlx4_en_do_multicast(priv, dev, mdev);
1268 out:
1269 	mutex_unlock(&mdev->state_lock);
1270 }
1271 
1272 #ifdef CONFIG_NET_POLL_CONTROLLER
1273 static void mlx4_en_netpoll(struct net_device *dev)
1274 {
1275 	struct mlx4_en_priv *priv = netdev_priv(dev);
1276 	struct mlx4_en_cq *cq;
1277 	int i;
1278 
1279 	for (i = 0; i < priv->rx_ring_num; i++) {
1280 		cq = priv->rx_cq[i];
1281 		napi_schedule(&cq->napi);
1282 	}
1283 }
1284 #endif
1285 
1286 static void mlx4_en_tx_timeout(struct net_device *dev)
1287 {
1288 	struct mlx4_en_priv *priv = netdev_priv(dev);
1289 	struct mlx4_en_dev *mdev = priv->mdev;
1290 	int i;
1291 
1292 	if (netif_msg_timer(priv))
1293 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1294 
1295 	for (i = 0; i < priv->tx_ring_num; i++) {
1296 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1297 			continue;
1298 		en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1299 			i, priv->tx_ring[i]->qpn, priv->tx_ring[i]->cqn,
1300 			priv->tx_ring[i]->cons, priv->tx_ring[i]->prod);
1301 	}
1302 
1303 	priv->port_stats.tx_timeout++;
1304 	en_dbg(DRV, priv, "Scheduling watchdog\n");
1305 	queue_work(mdev->workqueue, &priv->watchdog_task);
1306 }
1307 
1308 
1309 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
1310 {
1311 	struct mlx4_en_priv *priv = netdev_priv(dev);
1312 
1313 	spin_lock_bh(&priv->stats_lock);
1314 	memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
1315 	spin_unlock_bh(&priv->stats_lock);
1316 
1317 	return &priv->ret_stats;
1318 }
1319 
1320 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1321 {
1322 	struct mlx4_en_cq *cq;
1323 	int i;
1324 
1325 	/* If we haven't received a specific coalescing setting
1326 	 * (module param), we set the moderation parameters as follows:
1327 	 * - moder_cnt is set to the number of mtu sized packets to
1328 	 *   satisfy our coalescing target.
1329 	 * - moder_time is set to a fixed value.
1330 	 */
1331 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1332 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1333 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1334 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1335 	en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1336 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1337 
1338 	/* Setup cq moderation params */
1339 	for (i = 0; i < priv->rx_ring_num; i++) {
1340 		cq = priv->rx_cq[i];
1341 		cq->moder_cnt = priv->rx_frames;
1342 		cq->moder_time = priv->rx_usecs;
1343 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1344 		priv->last_moder_packets[i] = 0;
1345 		priv->last_moder_bytes[i] = 0;
1346 	}
1347 
1348 	for (i = 0; i < priv->tx_ring_num; i++) {
1349 		cq = priv->tx_cq[i];
1350 		cq->moder_cnt = priv->tx_frames;
1351 		cq->moder_time = priv->tx_usecs;
1352 	}
1353 
1354 	/* Reset auto-moderation params */
1355 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1356 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1357 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1358 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1359 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1360 	priv->adaptive_rx_coal = 1;
1361 	priv->last_moder_jiffies = 0;
1362 	priv->last_moder_tx_packets = 0;
1363 }
1364 
1365 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1366 {
1367 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1368 	struct mlx4_en_cq *cq;
1369 	unsigned long packets;
1370 	unsigned long rate;
1371 	unsigned long avg_pkt_size;
1372 	unsigned long rx_packets;
1373 	unsigned long rx_bytes;
1374 	unsigned long rx_pkt_diff;
1375 	int moder_time;
1376 	int ring, err;
1377 
1378 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1379 		return;
1380 
1381 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1382 		spin_lock_bh(&priv->stats_lock);
1383 		rx_packets = priv->rx_ring[ring]->packets;
1384 		rx_bytes = priv->rx_ring[ring]->bytes;
1385 		spin_unlock_bh(&priv->stats_lock);
1386 
1387 		rx_pkt_diff = ((unsigned long) (rx_packets -
1388 				priv->last_moder_packets[ring]));
1389 		packets = rx_pkt_diff;
1390 		rate = packets * HZ / period;
1391 		avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1392 				priv->last_moder_bytes[ring])) / packets : 0;
1393 
1394 		/* Apply auto-moderation only when packet rate
1395 		 * exceeds a rate that it matters */
1396 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1397 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1398 			if (rate < priv->pkt_rate_low)
1399 				moder_time = priv->rx_usecs_low;
1400 			else if (rate > priv->pkt_rate_high)
1401 				moder_time = priv->rx_usecs_high;
1402 			else
1403 				moder_time = (rate - priv->pkt_rate_low) *
1404 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1405 					(priv->pkt_rate_high - priv->pkt_rate_low) +
1406 					priv->rx_usecs_low;
1407 		} else {
1408 			moder_time = priv->rx_usecs_low;
1409 		}
1410 
1411 		if (moder_time != priv->last_moder_time[ring]) {
1412 			priv->last_moder_time[ring] = moder_time;
1413 			cq = priv->rx_cq[ring];
1414 			cq->moder_time = moder_time;
1415 			cq->moder_cnt = priv->rx_frames;
1416 			err = mlx4_en_set_cq_moder(priv, cq);
1417 			if (err)
1418 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1419 				       ring);
1420 		}
1421 		priv->last_moder_packets[ring] = rx_packets;
1422 		priv->last_moder_bytes[ring] = rx_bytes;
1423 	}
1424 
1425 	priv->last_moder_jiffies = jiffies;
1426 }
1427 
1428 static void mlx4_en_do_get_stats(struct work_struct *work)
1429 {
1430 	struct delayed_work *delay = to_delayed_work(work);
1431 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1432 						 stats_task);
1433 	struct mlx4_en_dev *mdev = priv->mdev;
1434 	int err;
1435 
1436 	mutex_lock(&mdev->state_lock);
1437 	if (mdev->device_up) {
1438 		if (priv->port_up) {
1439 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1440 			if (err)
1441 				en_dbg(HW, priv, "Could not update stats\n");
1442 
1443 			mlx4_en_auto_moderation(priv);
1444 		}
1445 
1446 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1447 	}
1448 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1449 		mlx4_en_do_set_mac(priv, priv->current_mac);
1450 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1451 	}
1452 	mutex_unlock(&mdev->state_lock);
1453 }
1454 
1455 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1456  * periodically
1457  */
1458 static void mlx4_en_service_task(struct work_struct *work)
1459 {
1460 	struct delayed_work *delay = to_delayed_work(work);
1461 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1462 						 service_task);
1463 	struct mlx4_en_dev *mdev = priv->mdev;
1464 
1465 	mutex_lock(&mdev->state_lock);
1466 	if (mdev->device_up) {
1467 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1468 			mlx4_en_ptp_overflow_check(mdev);
1469 
1470 		mlx4_en_recover_from_oom(priv);
1471 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1472 				   SERVICE_TASK_DELAY);
1473 	}
1474 	mutex_unlock(&mdev->state_lock);
1475 }
1476 
1477 static void mlx4_en_linkstate(struct work_struct *work)
1478 {
1479 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1480 						 linkstate_task);
1481 	struct mlx4_en_dev *mdev = priv->mdev;
1482 	int linkstate = priv->link_state;
1483 
1484 	mutex_lock(&mdev->state_lock);
1485 	/* If observable port state changed set carrier state and
1486 	 * report to system log */
1487 	if (priv->last_link_state != linkstate) {
1488 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1489 			en_info(priv, "Link Down\n");
1490 			netif_carrier_off(priv->dev);
1491 		} else {
1492 			en_info(priv, "Link Up\n");
1493 			netif_carrier_on(priv->dev);
1494 		}
1495 	}
1496 	priv->last_link_state = linkstate;
1497 	mutex_unlock(&mdev->state_lock);
1498 }
1499 
1500 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1501 {
1502 	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1503 	int numa_node = priv->mdev->dev->numa_node;
1504 
1505 	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1506 		return -ENOMEM;
1507 
1508 	cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1509 			ring->affinity_mask);
1510 	return 0;
1511 }
1512 
1513 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1514 {
1515 	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1516 }
1517 
1518 int mlx4_en_start_port(struct net_device *dev)
1519 {
1520 	struct mlx4_en_priv *priv = netdev_priv(dev);
1521 	struct mlx4_en_dev *mdev = priv->mdev;
1522 	struct mlx4_en_cq *cq;
1523 	struct mlx4_en_tx_ring *tx_ring;
1524 	int rx_index = 0;
1525 	int tx_index = 0;
1526 	int err = 0;
1527 	int i;
1528 	int j;
1529 	u8 mc_list[16] = {0};
1530 
1531 	if (priv->port_up) {
1532 		en_dbg(DRV, priv, "start port called while port already up\n");
1533 		return 0;
1534 	}
1535 
1536 	INIT_LIST_HEAD(&priv->mc_list);
1537 	INIT_LIST_HEAD(&priv->curr_list);
1538 	INIT_LIST_HEAD(&priv->ethtool_list);
1539 	memset(&priv->ethtool_rules[0], 0,
1540 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1541 
1542 	/* Calculate Rx buf size */
1543 	dev->mtu = min(dev->mtu, priv->max_mtu);
1544 	mlx4_en_calc_rx_buf(dev);
1545 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1546 
1547 	/* Configure rx cq's and rings */
1548 	err = mlx4_en_activate_rx_rings(priv);
1549 	if (err) {
1550 		en_err(priv, "Failed to activate RX rings\n");
1551 		return err;
1552 	}
1553 	for (i = 0; i < priv->rx_ring_num; i++) {
1554 		cq = priv->rx_cq[i];
1555 
1556 		mlx4_en_cq_init_lock(cq);
1557 
1558 		err = mlx4_en_init_affinity_hint(priv, i);
1559 		if (err) {
1560 			en_err(priv, "Failed preparing IRQ affinity hint\n");
1561 			goto cq_err;
1562 		}
1563 
1564 		err = mlx4_en_activate_cq(priv, cq, i);
1565 		if (err) {
1566 			en_err(priv, "Failed activating Rx CQ\n");
1567 			mlx4_en_free_affinity_hint(priv, i);
1568 			goto cq_err;
1569 		}
1570 
1571 		for (j = 0; j < cq->size; j++) {
1572 			struct mlx4_cqe *cqe = NULL;
1573 
1574 			cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1575 			      priv->cqe_factor;
1576 			cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1577 		}
1578 
1579 		err = mlx4_en_set_cq_moder(priv, cq);
1580 		if (err) {
1581 			en_err(priv, "Failed setting cq moderation parameters\n");
1582 			mlx4_en_deactivate_cq(priv, cq);
1583 			mlx4_en_free_affinity_hint(priv, i);
1584 			goto cq_err;
1585 		}
1586 		mlx4_en_arm_cq(priv, cq);
1587 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1588 		++rx_index;
1589 	}
1590 
1591 	/* Set qp number */
1592 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1593 	err = mlx4_en_get_qp(priv);
1594 	if (err) {
1595 		en_err(priv, "Failed getting eth qp\n");
1596 		goto cq_err;
1597 	}
1598 	mdev->mac_removed[priv->port] = 0;
1599 
1600 	priv->counter_index =
1601 			mlx4_get_default_counter_index(mdev->dev, priv->port);
1602 
1603 	err = mlx4_en_config_rss_steer(priv);
1604 	if (err) {
1605 		en_err(priv, "Failed configuring rss steering\n");
1606 		goto mac_err;
1607 	}
1608 
1609 	err = mlx4_en_create_drop_qp(priv);
1610 	if (err)
1611 		goto rss_err;
1612 
1613 	/* Configure tx cq's and rings */
1614 	for (i = 0; i < priv->tx_ring_num; i++) {
1615 		/* Configure cq */
1616 		cq = priv->tx_cq[i];
1617 		err = mlx4_en_activate_cq(priv, cq, i);
1618 		if (err) {
1619 			en_err(priv, "Failed allocating Tx CQ\n");
1620 			goto tx_err;
1621 		}
1622 		err = mlx4_en_set_cq_moder(priv, cq);
1623 		if (err) {
1624 			en_err(priv, "Failed setting cq moderation parameters\n");
1625 			mlx4_en_deactivate_cq(priv, cq);
1626 			goto tx_err;
1627 		}
1628 		en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1629 		cq->buf->wqe_index = cpu_to_be16(0xffff);
1630 
1631 		/* Configure ring */
1632 		tx_ring = priv->tx_ring[i];
1633 		err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1634 			i / priv->num_tx_rings_p_up);
1635 		if (err) {
1636 			en_err(priv, "Failed allocating Tx ring\n");
1637 			mlx4_en_deactivate_cq(priv, cq);
1638 			goto tx_err;
1639 		}
1640 		tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1641 
1642 		/* Arm CQ for TX completions */
1643 		mlx4_en_arm_cq(priv, cq);
1644 
1645 		/* Set initial ownership of all Tx TXBBs to SW (1) */
1646 		for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1647 			*((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1648 		++tx_index;
1649 	}
1650 
1651 	/* Configure port */
1652 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1653 				    priv->rx_skb_size + ETH_FCS_LEN,
1654 				    priv->prof->tx_pause,
1655 				    priv->prof->tx_ppp,
1656 				    priv->prof->rx_pause,
1657 				    priv->prof->rx_ppp);
1658 	if (err) {
1659 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1660 		       priv->port, err);
1661 		goto tx_err;
1662 	}
1663 	/* Set default qp number */
1664 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1665 	if (err) {
1666 		en_err(priv, "Failed setting default qp numbers\n");
1667 		goto tx_err;
1668 	}
1669 
1670 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1671 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1672 		if (err) {
1673 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1674 			       err);
1675 			goto tx_err;
1676 		}
1677 	}
1678 
1679 	/* Init port */
1680 	en_dbg(HW, priv, "Initializing port\n");
1681 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1682 	if (err) {
1683 		en_err(priv, "Failed Initializing port\n");
1684 		goto tx_err;
1685 	}
1686 
1687 	/* Attach rx QP to bradcast address */
1688 	eth_broadcast_addr(&mc_list[10]);
1689 	mc_list[5] = priv->port; /* needed for B0 steering support */
1690 	if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1691 				  priv->port, 0, MLX4_PROT_ETH,
1692 				  &priv->broadcast_id))
1693 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1694 
1695 	/* Must redo promiscuous mode setup. */
1696 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1697 
1698 	/* Schedule multicast task to populate multicast list */
1699 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1700 
1701 #ifdef CONFIG_MLX4_EN_VXLAN
1702 	if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1703 		vxlan_get_rx_port(dev);
1704 #endif
1705 	priv->port_up = true;
1706 	netif_tx_start_all_queues(dev);
1707 	netif_device_attach(dev);
1708 
1709 	return 0;
1710 
1711 tx_err:
1712 	while (tx_index--) {
1713 		mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1714 		mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1715 	}
1716 	mlx4_en_destroy_drop_qp(priv);
1717 rss_err:
1718 	mlx4_en_release_rss_steer(priv);
1719 mac_err:
1720 	mlx4_en_put_qp(priv);
1721 cq_err:
1722 	while (rx_index--) {
1723 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1724 		mlx4_en_free_affinity_hint(priv, rx_index);
1725 	}
1726 	for (i = 0; i < priv->rx_ring_num; i++)
1727 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1728 
1729 	return err; /* need to close devices */
1730 }
1731 
1732 
1733 void mlx4_en_stop_port(struct net_device *dev, int detach)
1734 {
1735 	struct mlx4_en_priv *priv = netdev_priv(dev);
1736 	struct mlx4_en_dev *mdev = priv->mdev;
1737 	struct mlx4_en_mc_list *mclist, *tmp;
1738 	struct ethtool_flow_id *flow, *tmp_flow;
1739 	int i;
1740 	u8 mc_list[16] = {0};
1741 
1742 	if (!priv->port_up) {
1743 		en_dbg(DRV, priv, "stop port called while port already down\n");
1744 		return;
1745 	}
1746 
1747 	/* close port*/
1748 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1749 
1750 	/* Synchronize with tx routine */
1751 	netif_tx_lock_bh(dev);
1752 	if (detach)
1753 		netif_device_detach(dev);
1754 	netif_tx_stop_all_queues(dev);
1755 	netif_tx_unlock_bh(dev);
1756 
1757 	netif_tx_disable(dev);
1758 
1759 	/* Set port as not active */
1760 	priv->port_up = false;
1761 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1762 
1763 	/* Promsicuous mode */
1764 	if (mdev->dev->caps.steering_mode ==
1765 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1766 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1767 				 MLX4_EN_FLAG_MC_PROMISC);
1768 		mlx4_flow_steer_promisc_remove(mdev->dev,
1769 					       priv->port,
1770 					       MLX4_FS_ALL_DEFAULT);
1771 		mlx4_flow_steer_promisc_remove(mdev->dev,
1772 					       priv->port,
1773 					       MLX4_FS_MC_DEFAULT);
1774 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1775 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1776 
1777 		/* Disable promiscouos mode */
1778 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1779 					    priv->port);
1780 
1781 		/* Disable Multicast promisc */
1782 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1783 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1784 						      priv->port);
1785 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1786 		}
1787 	}
1788 
1789 	/* Detach All multicasts */
1790 	eth_broadcast_addr(&mc_list[10]);
1791 	mc_list[5] = priv->port; /* needed for B0 steering support */
1792 	mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1793 			      MLX4_PROT_ETH, priv->broadcast_id);
1794 	list_for_each_entry(mclist, &priv->curr_list, list) {
1795 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1796 		mc_list[5] = priv->port;
1797 		mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1798 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1799 		if (mclist->tunnel_reg_id)
1800 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1801 	}
1802 	mlx4_en_clear_list(dev);
1803 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1804 		list_del(&mclist->list);
1805 		kfree(mclist);
1806 	}
1807 
1808 	/* Flush multicast filter */
1809 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1810 
1811 	/* Remove flow steering rules for the port*/
1812 	if (mdev->dev->caps.steering_mode ==
1813 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1814 		ASSERT_RTNL();
1815 		list_for_each_entry_safe(flow, tmp_flow,
1816 					 &priv->ethtool_list, list) {
1817 			mlx4_flow_detach(mdev->dev, flow->id);
1818 			list_del(&flow->list);
1819 		}
1820 	}
1821 
1822 	mlx4_en_destroy_drop_qp(priv);
1823 
1824 	/* Free TX Rings */
1825 	for (i = 0; i < priv->tx_ring_num; i++) {
1826 		mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1827 		mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1828 	}
1829 	msleep(10);
1830 
1831 	for (i = 0; i < priv->tx_ring_num; i++)
1832 		mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1833 
1834 	/* Free RSS qps */
1835 	mlx4_en_release_rss_steer(priv);
1836 
1837 	/* Unregister Mac address for the port */
1838 	mlx4_en_put_qp(priv);
1839 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1840 		mdev->mac_removed[priv->port] = 1;
1841 
1842 	/* Free RX Rings */
1843 	for (i = 0; i < priv->rx_ring_num; i++) {
1844 		struct mlx4_en_cq *cq = priv->rx_cq[i];
1845 
1846 		local_bh_disable();
1847 		while (!mlx4_en_cq_lock_napi(cq)) {
1848 			pr_info("CQ %d locked\n", i);
1849 			mdelay(1);
1850 		}
1851 		local_bh_enable();
1852 
1853 		napi_synchronize(&cq->napi);
1854 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1855 		mlx4_en_deactivate_cq(priv, cq);
1856 
1857 		mlx4_en_free_affinity_hint(priv, i);
1858 	}
1859 }
1860 
1861 static void mlx4_en_restart(struct work_struct *work)
1862 {
1863 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1864 						 watchdog_task);
1865 	struct mlx4_en_dev *mdev = priv->mdev;
1866 	struct net_device *dev = priv->dev;
1867 
1868 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1869 
1870 	mutex_lock(&mdev->state_lock);
1871 	if (priv->port_up) {
1872 		mlx4_en_stop_port(dev, 1);
1873 		if (mlx4_en_start_port(dev))
1874 			en_err(priv, "Failed restarting port %d\n", priv->port);
1875 	}
1876 	mutex_unlock(&mdev->state_lock);
1877 }
1878 
1879 static void mlx4_en_clear_stats(struct net_device *dev)
1880 {
1881 	struct mlx4_en_priv *priv = netdev_priv(dev);
1882 	struct mlx4_en_dev *mdev = priv->mdev;
1883 	int i;
1884 
1885 	if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1886 		en_dbg(HW, priv, "Failed dumping statistics\n");
1887 
1888 	memset(&priv->stats, 0, sizeof(priv->stats));
1889 	memset(&priv->pstats, 0, sizeof(priv->pstats));
1890 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1891 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1892 	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
1893 	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
1894 	memset(&priv->rx_priority_flowstats, 0,
1895 	       sizeof(priv->rx_priority_flowstats));
1896 	memset(&priv->tx_priority_flowstats, 0,
1897 	       sizeof(priv->tx_priority_flowstats));
1898 	memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
1899 
1900 	for (i = 0; i < priv->tx_ring_num; i++) {
1901 		priv->tx_ring[i]->bytes = 0;
1902 		priv->tx_ring[i]->packets = 0;
1903 		priv->tx_ring[i]->tx_csum = 0;
1904 	}
1905 	for (i = 0; i < priv->rx_ring_num; i++) {
1906 		priv->rx_ring[i]->bytes = 0;
1907 		priv->rx_ring[i]->packets = 0;
1908 		priv->rx_ring[i]->csum_ok = 0;
1909 		priv->rx_ring[i]->csum_none = 0;
1910 		priv->rx_ring[i]->csum_complete = 0;
1911 	}
1912 }
1913 
1914 static int mlx4_en_open(struct net_device *dev)
1915 {
1916 	struct mlx4_en_priv *priv = netdev_priv(dev);
1917 	struct mlx4_en_dev *mdev = priv->mdev;
1918 	int err = 0;
1919 
1920 	mutex_lock(&mdev->state_lock);
1921 
1922 	if (!mdev->device_up) {
1923 		en_err(priv, "Cannot open - device down/disabled\n");
1924 		err = -EBUSY;
1925 		goto out;
1926 	}
1927 
1928 	/* Reset HW statistics and SW counters */
1929 	mlx4_en_clear_stats(dev);
1930 
1931 	err = mlx4_en_start_port(dev);
1932 	if (err)
1933 		en_err(priv, "Failed starting port:%d\n", priv->port);
1934 
1935 out:
1936 	mutex_unlock(&mdev->state_lock);
1937 	return err;
1938 }
1939 
1940 
1941 static int mlx4_en_close(struct net_device *dev)
1942 {
1943 	struct mlx4_en_priv *priv = netdev_priv(dev);
1944 	struct mlx4_en_dev *mdev = priv->mdev;
1945 
1946 	en_dbg(IFDOWN, priv, "Close port called\n");
1947 
1948 	mutex_lock(&mdev->state_lock);
1949 
1950 	mlx4_en_stop_port(dev, 0);
1951 	netif_carrier_off(dev);
1952 
1953 	mutex_unlock(&mdev->state_lock);
1954 	return 0;
1955 }
1956 
1957 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1958 {
1959 	int i;
1960 
1961 #ifdef CONFIG_RFS_ACCEL
1962 	priv->dev->rx_cpu_rmap = NULL;
1963 #endif
1964 
1965 	for (i = 0; i < priv->tx_ring_num; i++) {
1966 		if (priv->tx_ring && priv->tx_ring[i])
1967 			mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1968 		if (priv->tx_cq && priv->tx_cq[i])
1969 			mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1970 	}
1971 
1972 	for (i = 0; i < priv->rx_ring_num; i++) {
1973 		if (priv->rx_ring[i])
1974 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1975 				priv->prof->rx_ring_size, priv->stride);
1976 		if (priv->rx_cq[i])
1977 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1978 	}
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 node;
1987 
1988 	/* Create tx Rings */
1989 	for (i = 0; i < priv->tx_ring_num; i++) {
1990 		node = cpu_to_node(i % num_online_cpus());
1991 		if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1992 				      prof->tx_ring_size, i, TX, node))
1993 			goto err;
1994 
1995 		if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
1996 					   prof->tx_ring_size, TXBB_SIZE,
1997 					   node, i))
1998 			goto err;
1999 	}
2000 
2001 	/* Create rx Rings */
2002 	for (i = 0; i < priv->rx_ring_num; i++) {
2003 		node = cpu_to_node(i % num_online_cpus());
2004 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2005 				      prof->rx_ring_size, i, RX, node))
2006 			goto err;
2007 
2008 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2009 					   prof->rx_ring_size, priv->stride,
2010 					   node))
2011 			goto err;
2012 	}
2013 
2014 #ifdef CONFIG_RFS_ACCEL
2015 	priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2016 #endif
2017 
2018 	return 0;
2019 
2020 err:
2021 	en_err(priv, "Failed to allocate NIC resources\n");
2022 	for (i = 0; i < priv->rx_ring_num; i++) {
2023 		if (priv->rx_ring[i])
2024 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2025 						prof->rx_ring_size,
2026 						priv->stride);
2027 		if (priv->rx_cq[i])
2028 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2029 	}
2030 	for (i = 0; i < priv->tx_ring_num; i++) {
2031 		if (priv->tx_ring[i])
2032 			mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
2033 		if (priv->tx_cq[i])
2034 			mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
2035 	}
2036 	return -ENOMEM;
2037 }
2038 
2039 
2040 void mlx4_en_destroy_netdev(struct net_device *dev)
2041 {
2042 	struct mlx4_en_priv *priv = netdev_priv(dev);
2043 	struct mlx4_en_dev *mdev = priv->mdev;
2044 
2045 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2046 
2047 	/* Unregister device - this will close the port if it was up */
2048 	if (priv->registered)
2049 		unregister_netdev(dev);
2050 
2051 	if (priv->allocated)
2052 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2053 
2054 	cancel_delayed_work(&priv->stats_task);
2055 	cancel_delayed_work(&priv->service_task);
2056 	/* flush any pending task for this netdev */
2057 	flush_workqueue(mdev->workqueue);
2058 
2059 	/* Detach the netdev so tasks would not attempt to access it */
2060 	mutex_lock(&mdev->state_lock);
2061 	mdev->pndev[priv->port] = NULL;
2062 	mdev->upper[priv->port] = NULL;
2063 	mutex_unlock(&mdev->state_lock);
2064 
2065 	mlx4_en_free_resources(priv);
2066 
2067 	kfree(priv->tx_ring);
2068 	kfree(priv->tx_cq);
2069 
2070 	free_netdev(dev);
2071 }
2072 
2073 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2074 {
2075 	struct mlx4_en_priv *priv = netdev_priv(dev);
2076 	struct mlx4_en_dev *mdev = priv->mdev;
2077 	int err = 0;
2078 
2079 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2080 		 dev->mtu, new_mtu);
2081 
2082 	if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
2083 		en_err(priv, "Bad MTU size:%d.\n", new_mtu);
2084 		return -EPERM;
2085 	}
2086 	dev->mtu = new_mtu;
2087 
2088 	if (netif_running(dev)) {
2089 		mutex_lock(&mdev->state_lock);
2090 		if (!mdev->device_up) {
2091 			/* NIC is probably restarting - let watchdog task reset
2092 			 * the port */
2093 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2094 		} else {
2095 			mlx4_en_stop_port(dev, 1);
2096 			err = mlx4_en_start_port(dev);
2097 			if (err) {
2098 				en_err(priv, "Failed restarting port:%d\n",
2099 					 priv->port);
2100 				queue_work(mdev->workqueue, &priv->watchdog_task);
2101 			}
2102 		}
2103 		mutex_unlock(&mdev->state_lock);
2104 	}
2105 	return 0;
2106 }
2107 
2108 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2109 {
2110 	struct mlx4_en_priv *priv = netdev_priv(dev);
2111 	struct mlx4_en_dev *mdev = priv->mdev;
2112 	struct hwtstamp_config config;
2113 
2114 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2115 		return -EFAULT;
2116 
2117 	/* reserved for future extensions */
2118 	if (config.flags)
2119 		return -EINVAL;
2120 
2121 	/* device doesn't support time stamping */
2122 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2123 		return -EINVAL;
2124 
2125 	/* TX HW timestamp */
2126 	switch (config.tx_type) {
2127 	case HWTSTAMP_TX_OFF:
2128 	case HWTSTAMP_TX_ON:
2129 		break;
2130 	default:
2131 		return -ERANGE;
2132 	}
2133 
2134 	/* RX HW timestamp */
2135 	switch (config.rx_filter) {
2136 	case HWTSTAMP_FILTER_NONE:
2137 		break;
2138 	case HWTSTAMP_FILTER_ALL:
2139 	case HWTSTAMP_FILTER_SOME:
2140 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2141 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2142 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2143 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2144 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2145 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2146 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2147 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2148 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2149 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2150 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2151 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2152 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2153 		break;
2154 	default:
2155 		return -ERANGE;
2156 	}
2157 
2158 	if (mlx4_en_reset_config(dev, config, dev->features)) {
2159 		config.tx_type = HWTSTAMP_TX_OFF;
2160 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2161 	}
2162 
2163 	return copy_to_user(ifr->ifr_data, &config,
2164 			    sizeof(config)) ? -EFAULT : 0;
2165 }
2166 
2167 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2168 {
2169 	struct mlx4_en_priv *priv = netdev_priv(dev);
2170 
2171 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2172 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2173 }
2174 
2175 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2176 {
2177 	switch (cmd) {
2178 	case SIOCSHWTSTAMP:
2179 		return mlx4_en_hwtstamp_set(dev, ifr);
2180 	case SIOCGHWTSTAMP:
2181 		return mlx4_en_hwtstamp_get(dev, ifr);
2182 	default:
2183 		return -EOPNOTSUPP;
2184 	}
2185 }
2186 
2187 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2188 					      netdev_features_t features)
2189 {
2190 	struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2191 	struct mlx4_en_dev *mdev = en_priv->mdev;
2192 
2193 	/* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2194 	 * enable/disable make sure S-TAG flag is always in same state as
2195 	 * C-TAG.
2196 	 */
2197 	if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2198 	    !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2199 		features |= NETIF_F_HW_VLAN_STAG_RX;
2200 	else
2201 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2202 
2203 	return features;
2204 }
2205 
2206 static int mlx4_en_set_features(struct net_device *netdev,
2207 		netdev_features_t features)
2208 {
2209 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2210 	bool reset = false;
2211 	int ret = 0;
2212 
2213 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2214 		en_info(priv, "Turn %s RX-FCS\n",
2215 			(features & NETIF_F_RXFCS) ? "ON" : "OFF");
2216 		reset = true;
2217 	}
2218 
2219 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2220 		u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2221 
2222 		en_info(priv, "Turn %s RX-ALL\n",
2223 			ignore_fcs_value ? "ON" : "OFF");
2224 		ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2225 					      priv->port, ignore_fcs_value);
2226 		if (ret)
2227 			return ret;
2228 	}
2229 
2230 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2231 		en_info(priv, "Turn %s RX vlan strip offload\n",
2232 			(features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2233 		reset = true;
2234 	}
2235 
2236 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2237 		en_info(priv, "Turn %s TX vlan strip offload\n",
2238 			(features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2239 
2240 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2241 		en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2242 			(features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2243 
2244 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2245 		en_info(priv, "Turn %s loopback\n",
2246 			(features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2247 		mlx4_en_update_loopback_state(netdev, features);
2248 	}
2249 
2250 	if (reset) {
2251 		ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2252 					   features);
2253 		if (ret)
2254 			return ret;
2255 	}
2256 
2257 	return 0;
2258 }
2259 
2260 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2261 {
2262 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2263 	struct mlx4_en_dev *mdev = en_priv->mdev;
2264 	u64 mac_u64 = mlx4_mac_to_u64(mac);
2265 
2266 	if (!is_valid_ether_addr(mac))
2267 		return -EINVAL;
2268 
2269 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2270 }
2271 
2272 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos)
2273 {
2274 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2275 	struct mlx4_en_dev *mdev = en_priv->mdev;
2276 
2277 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos);
2278 }
2279 
2280 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2281 			       int max_tx_rate)
2282 {
2283 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2284 	struct mlx4_en_dev *mdev = en_priv->mdev;
2285 
2286 	return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2287 				max_tx_rate);
2288 }
2289 
2290 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2291 {
2292 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2293 	struct mlx4_en_dev *mdev = en_priv->mdev;
2294 
2295 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2296 }
2297 
2298 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2299 {
2300 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2301 	struct mlx4_en_dev *mdev = en_priv->mdev;
2302 
2303 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2304 }
2305 
2306 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2307 {
2308 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2309 	struct mlx4_en_dev *mdev = en_priv->mdev;
2310 
2311 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2312 }
2313 
2314 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2315 				struct ifla_vf_stats *vf_stats)
2316 {
2317 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2318 	struct mlx4_en_dev *mdev = en_priv->mdev;
2319 
2320 	return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2321 }
2322 
2323 #define PORT_ID_BYTE_LEN 8
2324 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2325 				    struct netdev_phys_item_id *ppid)
2326 {
2327 	struct mlx4_en_priv *priv = netdev_priv(dev);
2328 	struct mlx4_dev *mdev = priv->mdev->dev;
2329 	int i;
2330 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2331 
2332 	if (!phys_port_id)
2333 		return -EOPNOTSUPP;
2334 
2335 	ppid->id_len = sizeof(phys_port_id);
2336 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2337 		ppid->id[i] =  phys_port_id & 0xff;
2338 		phys_port_id >>= 8;
2339 	}
2340 	return 0;
2341 }
2342 
2343 #ifdef CONFIG_MLX4_EN_VXLAN
2344 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2345 {
2346 	int ret;
2347 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2348 						 vxlan_add_task);
2349 
2350 	ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2351 	if (ret)
2352 		goto out;
2353 
2354 	ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2355 				  VXLAN_STEER_BY_OUTER_MAC, 1);
2356 out:
2357 	if (ret) {
2358 		en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2359 		return;
2360 	}
2361 
2362 	/* set offloads */
2363 	priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
2364 				      NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL;
2365 	priv->dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
2366 	priv->dev->features    |= NETIF_F_GSO_UDP_TUNNEL;
2367 }
2368 
2369 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2370 {
2371 	int ret;
2372 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2373 						 vxlan_del_task);
2374 	/* unset offloads */
2375 	priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
2376 				      NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL);
2377 	priv->dev->hw_features &= ~NETIF_F_GSO_UDP_TUNNEL;
2378 	priv->dev->features    &= ~NETIF_F_GSO_UDP_TUNNEL;
2379 
2380 	ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2381 				  VXLAN_STEER_BY_OUTER_MAC, 0);
2382 	if (ret)
2383 		en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2384 
2385 	priv->vxlan_port = 0;
2386 }
2387 
2388 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2389 				   sa_family_t sa_family, __be16 port)
2390 {
2391 	struct mlx4_en_priv *priv = netdev_priv(dev);
2392 	__be16 current_port;
2393 
2394 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2395 		return;
2396 
2397 	if (sa_family == AF_INET6)
2398 		return;
2399 
2400 	current_port = priv->vxlan_port;
2401 	if (current_port && current_port != port) {
2402 		en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2403 			ntohs(current_port), ntohs(port));
2404 		return;
2405 	}
2406 
2407 	priv->vxlan_port = port;
2408 	queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2409 }
2410 
2411 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2412 				   sa_family_t sa_family, __be16 port)
2413 {
2414 	struct mlx4_en_priv *priv = netdev_priv(dev);
2415 	__be16 current_port;
2416 
2417 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2418 		return;
2419 
2420 	if (sa_family == AF_INET6)
2421 		return;
2422 
2423 	current_port = priv->vxlan_port;
2424 	if (current_port != port) {
2425 		en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2426 		return;
2427 	}
2428 
2429 	queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2430 }
2431 
2432 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2433 						struct net_device *dev,
2434 						netdev_features_t features)
2435 {
2436 	features = vlan_features_check(skb, features);
2437 	return vxlan_features_check(skb, features);
2438 }
2439 #endif
2440 
2441 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2442 {
2443 	struct mlx4_en_priv *priv = netdev_priv(dev);
2444 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[queue_index];
2445 	struct mlx4_update_qp_params params;
2446 	int err;
2447 
2448 	if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2449 		return -EOPNOTSUPP;
2450 
2451 	/* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2452 	if (maxrate >> 12) {
2453 		params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2454 		params.rate_val  = maxrate / 1000;
2455 	} else if (maxrate) {
2456 		params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2457 		params.rate_val  = maxrate;
2458 	} else { /* zero serves to revoke the QP rate-limitation */
2459 		params.rate_unit = 0;
2460 		params.rate_val  = 0;
2461 	}
2462 
2463 	err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2464 			     &params);
2465 	return err;
2466 }
2467 
2468 static const struct net_device_ops mlx4_netdev_ops = {
2469 	.ndo_open		= mlx4_en_open,
2470 	.ndo_stop		= mlx4_en_close,
2471 	.ndo_start_xmit		= mlx4_en_xmit,
2472 	.ndo_select_queue	= mlx4_en_select_queue,
2473 	.ndo_get_stats		= mlx4_en_get_stats,
2474 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2475 	.ndo_set_mac_address	= mlx4_en_set_mac,
2476 	.ndo_validate_addr	= eth_validate_addr,
2477 	.ndo_change_mtu		= mlx4_en_change_mtu,
2478 	.ndo_do_ioctl		= mlx4_en_ioctl,
2479 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2480 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2481 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2482 #ifdef CONFIG_NET_POLL_CONTROLLER
2483 	.ndo_poll_controller	= mlx4_en_netpoll,
2484 #endif
2485 	.ndo_set_features	= mlx4_en_set_features,
2486 	.ndo_fix_features	= mlx4_en_fix_features,
2487 	.ndo_setup_tc		= mlx4_en_setup_tc,
2488 #ifdef CONFIG_RFS_ACCEL
2489 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2490 #endif
2491 #ifdef CONFIG_NET_RX_BUSY_POLL
2492 	.ndo_busy_poll		= mlx4_en_low_latency_recv,
2493 #endif
2494 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2495 #ifdef CONFIG_MLX4_EN_VXLAN
2496 	.ndo_add_vxlan_port	= mlx4_en_add_vxlan_port,
2497 	.ndo_del_vxlan_port	= mlx4_en_del_vxlan_port,
2498 	.ndo_features_check	= mlx4_en_features_check,
2499 #endif
2500 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2501 };
2502 
2503 static const struct net_device_ops mlx4_netdev_ops_master = {
2504 	.ndo_open		= mlx4_en_open,
2505 	.ndo_stop		= mlx4_en_close,
2506 	.ndo_start_xmit		= mlx4_en_xmit,
2507 	.ndo_select_queue	= mlx4_en_select_queue,
2508 	.ndo_get_stats		= mlx4_en_get_stats,
2509 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2510 	.ndo_set_mac_address	= mlx4_en_set_mac,
2511 	.ndo_validate_addr	= eth_validate_addr,
2512 	.ndo_change_mtu		= mlx4_en_change_mtu,
2513 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2514 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2515 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2516 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2517 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2518 	.ndo_set_vf_rate	= mlx4_en_set_vf_rate,
2519 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2520 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2521 	.ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2522 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2523 #ifdef CONFIG_NET_POLL_CONTROLLER
2524 	.ndo_poll_controller	= mlx4_en_netpoll,
2525 #endif
2526 	.ndo_set_features	= mlx4_en_set_features,
2527 	.ndo_fix_features	= mlx4_en_fix_features,
2528 	.ndo_setup_tc		= mlx4_en_setup_tc,
2529 #ifdef CONFIG_RFS_ACCEL
2530 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2531 #endif
2532 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2533 #ifdef CONFIG_MLX4_EN_VXLAN
2534 	.ndo_add_vxlan_port	= mlx4_en_add_vxlan_port,
2535 	.ndo_del_vxlan_port	= mlx4_en_del_vxlan_port,
2536 	.ndo_features_check	= mlx4_en_features_check,
2537 #endif
2538 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2539 };
2540 
2541 struct mlx4_en_bond {
2542 	struct work_struct work;
2543 	struct mlx4_en_priv *priv;
2544 	int is_bonded;
2545 	struct mlx4_port_map port_map;
2546 };
2547 
2548 static void mlx4_en_bond_work(struct work_struct *work)
2549 {
2550 	struct mlx4_en_bond *bond = container_of(work,
2551 						     struct mlx4_en_bond,
2552 						     work);
2553 	int err = 0;
2554 	struct mlx4_dev *dev = bond->priv->mdev->dev;
2555 
2556 	if (bond->is_bonded) {
2557 		if (!mlx4_is_bonded(dev)) {
2558 			err = mlx4_bond(dev);
2559 			if (err)
2560 				en_err(bond->priv, "Fail to bond device\n");
2561 		}
2562 		if (!err) {
2563 			err = mlx4_port_map_set(dev, &bond->port_map);
2564 			if (err)
2565 				en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2566 				       bond->port_map.port1,
2567 				       bond->port_map.port2,
2568 				       err);
2569 		}
2570 	} else if (mlx4_is_bonded(dev)) {
2571 		err = mlx4_unbond(dev);
2572 		if (err)
2573 			en_err(bond->priv, "Fail to unbond device\n");
2574 	}
2575 	dev_put(bond->priv->dev);
2576 	kfree(bond);
2577 }
2578 
2579 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2580 				   u8 v2p_p1, u8 v2p_p2)
2581 {
2582 	struct mlx4_en_bond *bond = NULL;
2583 
2584 	bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2585 	if (!bond)
2586 		return -ENOMEM;
2587 
2588 	INIT_WORK(&bond->work, mlx4_en_bond_work);
2589 	bond->priv = priv;
2590 	bond->is_bonded = is_bonded;
2591 	bond->port_map.port1 = v2p_p1;
2592 	bond->port_map.port2 = v2p_p2;
2593 	dev_hold(priv->dev);
2594 	queue_work(priv->mdev->workqueue, &bond->work);
2595 	return 0;
2596 }
2597 
2598 int mlx4_en_netdev_event(struct notifier_block *this,
2599 			 unsigned long event, void *ptr)
2600 {
2601 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2602 	u8 port = 0;
2603 	struct mlx4_en_dev *mdev;
2604 	struct mlx4_dev *dev;
2605 	int i, num_eth_ports = 0;
2606 	bool do_bond = true;
2607 	struct mlx4_en_priv *priv;
2608 	u8 v2p_port1 = 0;
2609 	u8 v2p_port2 = 0;
2610 
2611 	if (!net_eq(dev_net(ndev), &init_net))
2612 		return NOTIFY_DONE;
2613 
2614 	mdev = container_of(this, struct mlx4_en_dev, nb);
2615 	dev = mdev->dev;
2616 
2617 	/* Go into this mode only when two network devices set on two ports
2618 	 * of the same mlx4 device are slaves of the same bonding master
2619 	 */
2620 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2621 		++num_eth_ports;
2622 		if (!port && (mdev->pndev[i] == ndev))
2623 			port = i;
2624 		mdev->upper[i] = mdev->pndev[i] ?
2625 			netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2626 		/* condition not met: network device is a slave */
2627 		if (!mdev->upper[i])
2628 			do_bond = false;
2629 		if (num_eth_ports < 2)
2630 			continue;
2631 		/* condition not met: same master */
2632 		if (mdev->upper[i] != mdev->upper[i-1])
2633 			do_bond = false;
2634 	}
2635 	/* condition not met: 2 salves */
2636 	do_bond = (num_eth_ports ==  2) ? do_bond : false;
2637 
2638 	/* handle only events that come with enough info */
2639 	if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
2640 		return NOTIFY_DONE;
2641 
2642 	priv = netdev_priv(ndev);
2643 	if (do_bond) {
2644 		struct netdev_notifier_bonding_info *notifier_info = ptr;
2645 		struct netdev_bonding_info *bonding_info =
2646 			&notifier_info->bonding_info;
2647 
2648 		/* required mode 1, 2 or 4 */
2649 		if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
2650 		    (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
2651 		    (bonding_info->master.bond_mode != BOND_MODE_8023AD))
2652 			do_bond = false;
2653 
2654 		/* require exactly 2 slaves */
2655 		if (bonding_info->master.num_slaves != 2)
2656 			do_bond = false;
2657 
2658 		/* calc v2p */
2659 		if (do_bond) {
2660 			if (bonding_info->master.bond_mode ==
2661 			    BOND_MODE_ACTIVEBACKUP) {
2662 				/* in active-backup mode virtual ports are
2663 				 * mapped to the physical port of the active
2664 				 * slave */
2665 				if (bonding_info->slave.state ==
2666 				    BOND_STATE_BACKUP) {
2667 					if (port == 1) {
2668 						v2p_port1 = 2;
2669 						v2p_port2 = 2;
2670 					} else {
2671 						v2p_port1 = 1;
2672 						v2p_port2 = 1;
2673 					}
2674 				} else { /* BOND_STATE_ACTIVE */
2675 					if (port == 1) {
2676 						v2p_port1 = 1;
2677 						v2p_port2 = 1;
2678 					} else {
2679 						v2p_port1 = 2;
2680 						v2p_port2 = 2;
2681 					}
2682 				}
2683 			} else { /* Active-Active */
2684 				/* in active-active mode a virtual port is
2685 				 * mapped to the native physical port if and only
2686 				 * if the physical port is up */
2687 				__s8 link = bonding_info->slave.link;
2688 
2689 				if (port == 1)
2690 					v2p_port2 = 2;
2691 				else
2692 					v2p_port1 = 1;
2693 				if ((link == BOND_LINK_UP) ||
2694 				    (link == BOND_LINK_FAIL)) {
2695 					if (port == 1)
2696 						v2p_port1 = 1;
2697 					else
2698 						v2p_port2 = 2;
2699 				} else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
2700 					if (port == 1)
2701 						v2p_port1 = 2;
2702 					else
2703 						v2p_port2 = 1;
2704 				}
2705 			}
2706 		}
2707 	}
2708 
2709 	mlx4_en_queue_bond_work(priv, do_bond,
2710 				v2p_port1, v2p_port2);
2711 
2712 	return NOTIFY_DONE;
2713 }
2714 
2715 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
2716 				     struct mlx4_en_stats_bitmap *stats_bitmap,
2717 				     u8 rx_ppp, u8 rx_pause,
2718 				     u8 tx_ppp, u8 tx_pause)
2719 {
2720 	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
2721 
2722 	if (!mlx4_is_slave(dev) &&
2723 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
2724 		mutex_lock(&stats_bitmap->mutex);
2725 		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
2726 
2727 		if (rx_ppp)
2728 			bitmap_set(stats_bitmap->bitmap, last_i,
2729 				   NUM_FLOW_PRIORITY_STATS_RX);
2730 		last_i += NUM_FLOW_PRIORITY_STATS_RX;
2731 
2732 		if (rx_pause && !(rx_ppp))
2733 			bitmap_set(stats_bitmap->bitmap, last_i,
2734 				   NUM_FLOW_STATS_RX);
2735 		last_i += NUM_FLOW_STATS_RX;
2736 
2737 		if (tx_ppp)
2738 			bitmap_set(stats_bitmap->bitmap, last_i,
2739 				   NUM_FLOW_PRIORITY_STATS_TX);
2740 		last_i += NUM_FLOW_PRIORITY_STATS_TX;
2741 
2742 		if (tx_pause && !(tx_ppp))
2743 			bitmap_set(stats_bitmap->bitmap, last_i,
2744 				   NUM_FLOW_STATS_TX);
2745 		last_i += NUM_FLOW_STATS_TX;
2746 
2747 		mutex_unlock(&stats_bitmap->mutex);
2748 	}
2749 }
2750 
2751 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
2752 			      struct mlx4_en_stats_bitmap *stats_bitmap,
2753 			      u8 rx_ppp, u8 rx_pause,
2754 			      u8 tx_ppp, u8 tx_pause)
2755 {
2756 	int last_i = 0;
2757 
2758 	mutex_init(&stats_bitmap->mutex);
2759 	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
2760 
2761 	if (mlx4_is_slave(dev)) {
2762 		bitmap_set(stats_bitmap->bitmap, last_i +
2763 					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
2764 		bitmap_set(stats_bitmap->bitmap, last_i +
2765 					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
2766 		bitmap_set(stats_bitmap->bitmap, last_i +
2767 					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
2768 		bitmap_set(stats_bitmap->bitmap, last_i +
2769 					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
2770 		bitmap_set(stats_bitmap->bitmap, last_i +
2771 					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
2772 		bitmap_set(stats_bitmap->bitmap, last_i +
2773 					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
2774 	} else {
2775 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
2776 	}
2777 	last_i += NUM_MAIN_STATS;
2778 
2779 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
2780 	last_i += NUM_PORT_STATS;
2781 
2782 	if (mlx4_is_master(dev))
2783 		bitmap_set(stats_bitmap->bitmap, last_i,
2784 			   NUM_PF_STATS);
2785 	last_i += NUM_PF_STATS;
2786 
2787 	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
2788 					rx_ppp, rx_pause,
2789 					tx_ppp, tx_pause);
2790 	last_i += NUM_FLOW_STATS;
2791 
2792 	if (!mlx4_is_slave(dev))
2793 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
2794 }
2795 
2796 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2797 			struct mlx4_en_port_profile *prof)
2798 {
2799 	struct net_device *dev;
2800 	struct mlx4_en_priv *priv;
2801 	int i;
2802 	int err;
2803 	u64 mac_u64;
2804 
2805 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2806 				 MAX_TX_RINGS, MAX_RX_RINGS);
2807 	if (dev == NULL)
2808 		return -ENOMEM;
2809 
2810 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2811 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2812 
2813 	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
2814 	dev->dev_port = port - 1;
2815 
2816 	/*
2817 	 * Initialize driver private data
2818 	 */
2819 
2820 	priv = netdev_priv(dev);
2821 	memset(priv, 0, sizeof(struct mlx4_en_priv));
2822 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
2823 	spin_lock_init(&priv->stats_lock);
2824 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2825 	INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2826 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2827 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2828 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2829 #ifdef CONFIG_MLX4_EN_VXLAN
2830 	INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
2831 	INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
2832 #endif
2833 #ifdef CONFIG_RFS_ACCEL
2834 	INIT_LIST_HEAD(&priv->filters);
2835 	spin_lock_init(&priv->filters_lock);
2836 #endif
2837 
2838 	priv->dev = dev;
2839 	priv->mdev = mdev;
2840 	priv->ddev = &mdev->pdev->dev;
2841 	priv->prof = prof;
2842 	priv->port = port;
2843 	priv->port_up = false;
2844 	priv->flags = prof->flags;
2845 	priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
2846 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2847 			MLX4_WQE_CTRL_SOLICITED);
2848 	priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2849 	priv->tx_ring_num = prof->tx_ring_num;
2850 	priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
2851 	netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
2852 
2853 	priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2854 				GFP_KERNEL);
2855 	if (!priv->tx_ring) {
2856 		err = -ENOMEM;
2857 		goto out;
2858 	}
2859 	priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2860 			      GFP_KERNEL);
2861 	if (!priv->tx_cq) {
2862 		err = -ENOMEM;
2863 		goto out;
2864 	}
2865 	priv->rx_ring_num = prof->rx_ring_num;
2866 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2867 	priv->cqe_size = mdev->dev->caps.cqe_size;
2868 	priv->mac_index = -1;
2869 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
2870 #ifdef CONFIG_MLX4_EN_DCB
2871 	if (!mlx4_is_slave(priv->mdev->dev)) {
2872 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
2873 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2874 		} else {
2875 			en_info(priv, "enabling only PFC DCB ops\n");
2876 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2877 		}
2878 	}
2879 #endif
2880 
2881 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2882 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
2883 
2884 	/* Query for default mac and max mtu */
2885 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2886 
2887 	if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
2888 	    MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
2889 		priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
2890 
2891 	/* Set default MAC */
2892 	dev->addr_len = ETH_ALEN;
2893 	mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2894 	if (!is_valid_ether_addr(dev->dev_addr)) {
2895 		if (mlx4_is_slave(priv->mdev->dev)) {
2896 			eth_hw_addr_random(dev);
2897 			en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
2898 			mac_u64 = mlx4_mac_to_u64(dev->dev_addr);
2899 			mdev->dev->caps.def_mac[priv->port] = mac_u64;
2900 		} else {
2901 			en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2902 			       priv->port, dev->dev_addr);
2903 			err = -EINVAL;
2904 			goto out;
2905 		}
2906 	}
2907 
2908 	memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
2909 
2910 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2911 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2912 	err = mlx4_en_alloc_resources(priv);
2913 	if (err)
2914 		goto out;
2915 
2916 	/* Initialize time stamping config */
2917 	priv->hwtstamp_config.flags = 0;
2918 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
2919 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2920 
2921 	/* Allocate page for receive rings */
2922 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2923 				MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2924 	if (err) {
2925 		en_err(priv, "Failed to allocate page for rx qps\n");
2926 		goto out;
2927 	}
2928 	priv->allocated = 1;
2929 
2930 	/*
2931 	 * Initialize netdev entry points
2932 	 */
2933 	if (mlx4_is_master(priv->mdev->dev))
2934 		dev->netdev_ops = &mlx4_netdev_ops_master;
2935 	else
2936 		dev->netdev_ops = &mlx4_netdev_ops;
2937 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2938 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2939 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2940 
2941 	dev->ethtool_ops = &mlx4_en_ethtool_ops;
2942 
2943 	/*
2944 	 * Set driver features
2945 	 */
2946 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2947 	if (mdev->LSO_support)
2948 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2949 
2950 	dev->vlan_features = dev->hw_features;
2951 
2952 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2953 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2954 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2955 			NETIF_F_HW_VLAN_CTAG_FILTER;
2956 	dev->hw_features |= NETIF_F_LOOPBACK |
2957 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2958 
2959 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
2960 		dev->features |= NETIF_F_HW_VLAN_STAG_RX |
2961 			NETIF_F_HW_VLAN_STAG_FILTER;
2962 		dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
2963 	}
2964 
2965 	if (mlx4_is_slave(mdev->dev)) {
2966 		int phv;
2967 
2968 		err = get_phv_bit(mdev->dev, port, &phv);
2969 		if (!err && phv) {
2970 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
2971 			priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
2972 		}
2973 	} else {
2974 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
2975 		    !(mdev->dev->caps.flags2 &
2976 		      MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2977 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
2978 	}
2979 
2980 	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
2981 		dev->hw_features |= NETIF_F_RXFCS;
2982 
2983 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
2984 		dev->hw_features |= NETIF_F_RXALL;
2985 
2986 	if (mdev->dev->caps.steering_mode ==
2987 	    MLX4_STEERING_MODE_DEVICE_MANAGED &&
2988 	    mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
2989 		dev->hw_features |= NETIF_F_NTUPLE;
2990 
2991 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
2992 		dev->priv_flags |= IFF_UNICAST_FLT;
2993 
2994 	/* Setting a default hash function value */
2995 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
2996 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
2997 	} else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
2998 		priv->rss_hash_fn = ETH_RSS_HASH_XOR;
2999 	} else {
3000 		en_warn(priv,
3001 			"No RSS hash capabilities exposed, using Toeplitz\n");
3002 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3003 	}
3004 
3005 	mdev->pndev[port] = dev;
3006 	mdev->upper[port] = NULL;
3007 
3008 	netif_carrier_off(dev);
3009 	mlx4_en_set_default_moderation(priv);
3010 
3011 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
3012 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3013 
3014 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3015 
3016 	/* Configure port */
3017 	mlx4_en_calc_rx_buf(dev);
3018 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3019 				    priv->rx_skb_size + ETH_FCS_LEN,
3020 				    prof->tx_pause, prof->tx_ppp,
3021 				    prof->rx_pause, prof->rx_ppp);
3022 	if (err) {
3023 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3024 		       priv->port, err);
3025 		goto out;
3026 	}
3027 
3028 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3029 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3030 		if (err) {
3031 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3032 			       err);
3033 			goto out;
3034 		}
3035 	}
3036 
3037 	/* Init port */
3038 	en_warn(priv, "Initializing port\n");
3039 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
3040 	if (err) {
3041 		en_err(priv, "Failed Initializing port\n");
3042 		goto out;
3043 	}
3044 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3045 
3046 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3047 		queue_delayed_work(mdev->workqueue, &priv->service_task,
3048 				   SERVICE_TASK_DELAY);
3049 
3050 	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3051 				 mdev->profile.prof[priv->port].rx_ppp,
3052 				 mdev->profile.prof[priv->port].rx_pause,
3053 				 mdev->profile.prof[priv->port].tx_ppp,
3054 				 mdev->profile.prof[priv->port].tx_pause);
3055 
3056 	err = register_netdev(dev);
3057 	if (err) {
3058 		en_err(priv, "Netdev registration failed for port %d\n", port);
3059 		goto out;
3060 	}
3061 
3062 	priv->registered = 1;
3063 
3064 	return 0;
3065 
3066 out:
3067 	mlx4_en_destroy_netdev(dev);
3068 	return err;
3069 }
3070 
3071 int mlx4_en_reset_config(struct net_device *dev,
3072 			 struct hwtstamp_config ts_config,
3073 			 netdev_features_t features)
3074 {
3075 	struct mlx4_en_priv *priv = netdev_priv(dev);
3076 	struct mlx4_en_dev *mdev = priv->mdev;
3077 	int port_up = 0;
3078 	int err = 0;
3079 
3080 	if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3081 	    priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3082 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3083 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3084 		return 0; /* Nothing to change */
3085 
3086 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3087 	    (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3088 	    (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3089 		en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3090 		return -EINVAL;
3091 	}
3092 
3093 	mutex_lock(&mdev->state_lock);
3094 	if (priv->port_up) {
3095 		port_up = 1;
3096 		mlx4_en_stop_port(dev, 1);
3097 	}
3098 
3099 	mlx4_en_free_resources(priv);
3100 
3101 	en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
3102 		ts_config.rx_filter, !!(features & NETIF_F_HW_VLAN_CTAG_RX));
3103 
3104 	priv->hwtstamp_config.tx_type = ts_config.tx_type;
3105 	priv->hwtstamp_config.rx_filter = ts_config.rx_filter;
3106 
3107 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3108 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3109 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3110 		else
3111 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3112 	} else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3113 		/* RX time-stamping is OFF, update the RX vlan offload
3114 		 * to the latest wanted state
3115 		 */
3116 		if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3117 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3118 		else
3119 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3120 	}
3121 
3122 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3123 		if (features & NETIF_F_RXFCS)
3124 			dev->features |= NETIF_F_RXFCS;
3125 		else
3126 			dev->features &= ~NETIF_F_RXFCS;
3127 	}
3128 
3129 	/* RX vlan offload and RX time-stamping can't co-exist !
3130 	 * Regardless of the caller's choice,
3131 	 * Turn Off RX vlan offload in case of time-stamping is ON
3132 	 */
3133 	if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3134 		if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3135 			en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3136 		dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3137 	}
3138 
3139 	err = mlx4_en_alloc_resources(priv);
3140 	if (err) {
3141 		en_err(priv, "Failed reallocating port resources\n");
3142 		goto out;
3143 	}
3144 	if (port_up) {
3145 		err = mlx4_en_start_port(dev);
3146 		if (err)
3147 			en_err(priv, "Failed starting port\n");
3148 	}
3149 
3150 out:
3151 	mutex_unlock(&mdev->state_lock);
3152 	netdev_features_change(dev);
3153 	return err;
3154 }
3155