xref: /linux/drivers/net/ethernet/mellanox/mlx4/en_netdev.c (revision 0fdebc5ec2ca492d69df2d93a6a7abade4941aae)
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/bpf.h>
35 #include <linux/etherdevice.h>
36 #include <linux/filter.h>
37 #include <linux/tcp.h>
38 #include <linux/if_vlan.h>
39 #include <linux/delay.h>
40 #include <linux/slab.h>
41 #include <linux/hash.h>
42 #include <net/ip.h>
43 #include <net/vxlan.h>
44 #include <net/devlink.h>
45 
46 #include <linux/mlx4/driver.h>
47 #include <linux/mlx4/device.h>
48 #include <linux/mlx4/cmd.h>
49 #include <linux/mlx4/cq.h>
50 
51 #include "mlx4_en.h"
52 #include "en_port.h"
53 
54 #define MLX4_EN_MAX_XDP_MTU ((int)(PAGE_SIZE - ETH_HLEN - (2 * VLAN_HLEN) - \
55 				XDP_PACKET_HEADROOM -			    \
56 				SKB_DATA_ALIGN(sizeof(struct skb_shared_info))))
57 
58 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
59 {
60 	struct mlx4_en_priv *priv = netdev_priv(dev);
61 	int i;
62 	unsigned int offset = 0;
63 
64 	if (up && up != MLX4_EN_NUM_UP_HIGH)
65 		return -EINVAL;
66 
67 	netdev_set_num_tc(dev, up);
68 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
69 	/* Partition Tx queues evenly amongst UP's */
70 	for (i = 0; i < up; i++) {
71 		netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
72 		offset += priv->num_tx_rings_p_up;
73 	}
74 
75 #ifdef CONFIG_MLX4_EN_DCB
76 	if (!mlx4_is_slave(priv->mdev->dev)) {
77 		if (up) {
78 			if (priv->dcbx_cap)
79 				priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
80 		} else {
81 			priv->flags &= ~MLX4_EN_FLAG_DCB_ENABLED;
82 			priv->cee_config.pfc_state = false;
83 		}
84 	}
85 #endif /* CONFIG_MLX4_EN_DCB */
86 
87 	return 0;
88 }
89 
90 int mlx4_en_alloc_tx_queue_per_tc(struct net_device *dev, u8 tc)
91 {
92 	struct mlx4_en_priv *priv = netdev_priv(dev);
93 	struct mlx4_en_dev *mdev = priv->mdev;
94 	struct mlx4_en_port_profile new_prof;
95 	struct mlx4_en_priv *tmp;
96 	int total_count;
97 	int port_up = 0;
98 	int err = 0;
99 
100 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
101 	if (!tmp)
102 		return -ENOMEM;
103 
104 	mutex_lock(&mdev->state_lock);
105 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
106 	new_prof.num_up = (tc == 0) ? MLX4_EN_NUM_UP_LOW :
107 				      MLX4_EN_NUM_UP_HIGH;
108 	new_prof.tx_ring_num[TX] = new_prof.num_tx_rings_p_up *
109 				   new_prof.num_up;
110 	total_count = new_prof.tx_ring_num[TX] + new_prof.tx_ring_num[TX_XDP];
111 	if (total_count > MAX_TX_RINGS) {
112 		err = -EINVAL;
113 		en_err(priv,
114 		       "Total number of TX and XDP rings (%d) exceeds the maximum supported (%d)\n",
115 		       total_count, MAX_TX_RINGS);
116 		goto out;
117 	}
118 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
119 	if (err)
120 		goto out;
121 
122 	if (priv->port_up) {
123 		port_up = 1;
124 		mlx4_en_stop_port(dev, 1);
125 	}
126 
127 	mlx4_en_safe_replace_resources(priv, tmp);
128 	if (port_up) {
129 		err = mlx4_en_start_port(dev);
130 		if (err) {
131 			en_err(priv, "Failed starting port for setup TC\n");
132 			goto out;
133 		}
134 	}
135 
136 	err = mlx4_en_setup_tc(dev, tc);
137 out:
138 	mutex_unlock(&mdev->state_lock);
139 	kfree(tmp);
140 	return err;
141 }
142 
143 static int __mlx4_en_setup_tc(struct net_device *dev, enum tc_setup_type type,
144 			      void *type_data)
145 {
146 	struct tc_mqprio_qopt *mqprio = type_data;
147 
148 	if (type != TC_SETUP_QDISC_MQPRIO)
149 		return -EOPNOTSUPP;
150 
151 	if (mqprio->num_tc && mqprio->num_tc != MLX4_EN_NUM_UP_HIGH)
152 		return -EINVAL;
153 
154 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
155 
156 	return mlx4_en_alloc_tx_queue_per_tc(dev, mqprio->num_tc);
157 }
158 
159 #ifdef CONFIG_RFS_ACCEL
160 
161 struct mlx4_en_filter {
162 	struct list_head next;
163 	struct work_struct work;
164 
165 	u8     ip_proto;
166 	__be32 src_ip;
167 	__be32 dst_ip;
168 	__be16 src_port;
169 	__be16 dst_port;
170 
171 	int rxq_index;
172 	struct mlx4_en_priv *priv;
173 	u32 flow_id;			/* RFS infrastructure id */
174 	int id;				/* mlx4_en driver id */
175 	u64 reg_id;			/* Flow steering API id */
176 	u8 activated;			/* Used to prevent expiry before filter
177 					 * is attached
178 					 */
179 	struct hlist_node filter_chain;
180 };
181 
182 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
183 
184 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
185 {
186 	switch (ip_proto) {
187 	case IPPROTO_UDP:
188 		return MLX4_NET_TRANS_RULE_ID_UDP;
189 	case IPPROTO_TCP:
190 		return MLX4_NET_TRANS_RULE_ID_TCP;
191 	default:
192 		return MLX4_NET_TRANS_RULE_NUM;
193 	}
194 };
195 
196 /* Must not acquire state_lock, as its corresponding work_sync
197  * is done under it.
198  */
199 static void mlx4_en_filter_work(struct work_struct *work)
200 {
201 	struct mlx4_en_filter *filter = container_of(work,
202 						     struct mlx4_en_filter,
203 						     work);
204 	struct mlx4_en_priv *priv = filter->priv;
205 	struct mlx4_spec_list spec_tcp_udp = {
206 		.id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
207 		{
208 			.tcp_udp = {
209 				.dst_port = filter->dst_port,
210 				.dst_port_msk = (__force __be16)-1,
211 				.src_port = filter->src_port,
212 				.src_port_msk = (__force __be16)-1,
213 			},
214 		},
215 	};
216 	struct mlx4_spec_list spec_ip = {
217 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
218 		{
219 			.ipv4 = {
220 				.dst_ip = filter->dst_ip,
221 				.dst_ip_msk = (__force __be32)-1,
222 				.src_ip = filter->src_ip,
223 				.src_ip_msk = (__force __be32)-1,
224 			},
225 		},
226 	};
227 	struct mlx4_spec_list spec_eth = {
228 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
229 	};
230 	struct mlx4_net_trans_rule rule = {
231 		.list = LIST_HEAD_INIT(rule.list),
232 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
233 		.exclusive = 1,
234 		.allow_loopback = 1,
235 		.promisc_mode = MLX4_FS_REGULAR,
236 		.port = priv->port,
237 		.priority = MLX4_DOMAIN_RFS,
238 	};
239 	int rc;
240 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
241 
242 	if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
243 		en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
244 			filter->ip_proto);
245 		goto ignore;
246 	}
247 	list_add_tail(&spec_eth.list, &rule.list);
248 	list_add_tail(&spec_ip.list, &rule.list);
249 	list_add_tail(&spec_tcp_udp.list, &rule.list);
250 
251 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
252 	memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
253 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
254 
255 	filter->activated = 0;
256 
257 	if (filter->reg_id) {
258 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
259 		if (rc && rc != -ENOENT)
260 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
261 	}
262 
263 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
264 	if (rc)
265 		en_err(priv, "Error attaching flow. err = %d\n", rc);
266 
267 ignore:
268 	mlx4_en_filter_rfs_expire(priv);
269 
270 	filter->activated = 1;
271 }
272 
273 static inline struct hlist_head *
274 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
275 		   __be16 src_port, __be16 dst_port)
276 {
277 	unsigned long l;
278 	int bucket_idx;
279 
280 	l = (__force unsigned long)src_port |
281 	    ((__force unsigned long)dst_port << 2);
282 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
283 
284 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
285 
286 	return &priv->filter_hash[bucket_idx];
287 }
288 
289 static struct mlx4_en_filter *
290 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
291 		     __be32 dst_ip, u8 ip_proto, __be16 src_port,
292 		     __be16 dst_port, u32 flow_id)
293 {
294 	struct mlx4_en_filter *filter = NULL;
295 
296 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
297 	if (!filter)
298 		return NULL;
299 
300 	filter->priv = priv;
301 	filter->rxq_index = rxq_index;
302 	INIT_WORK(&filter->work, mlx4_en_filter_work);
303 
304 	filter->src_ip = src_ip;
305 	filter->dst_ip = dst_ip;
306 	filter->ip_proto = ip_proto;
307 	filter->src_port = src_port;
308 	filter->dst_port = dst_port;
309 
310 	filter->flow_id = flow_id;
311 
312 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
313 
314 	list_add_tail(&filter->next, &priv->filters);
315 	hlist_add_head(&filter->filter_chain,
316 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
317 					  dst_port));
318 
319 	return filter;
320 }
321 
322 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
323 {
324 	struct mlx4_en_priv *priv = filter->priv;
325 	int rc;
326 
327 	list_del(&filter->next);
328 
329 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
330 	if (rc && rc != -ENOENT)
331 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
332 
333 	kfree(filter);
334 }
335 
336 static inline struct mlx4_en_filter *
337 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
338 		    u8 ip_proto, __be16 src_port, __be16 dst_port)
339 {
340 	struct mlx4_en_filter *filter;
341 	struct mlx4_en_filter *ret = NULL;
342 
343 	hlist_for_each_entry(filter,
344 			     filter_hash_bucket(priv, src_ip, dst_ip,
345 						src_port, dst_port),
346 			     filter_chain) {
347 		if (filter->src_ip == src_ip &&
348 		    filter->dst_ip == dst_ip &&
349 		    filter->ip_proto == ip_proto &&
350 		    filter->src_port == src_port &&
351 		    filter->dst_port == dst_port) {
352 			ret = filter;
353 			break;
354 		}
355 	}
356 
357 	return ret;
358 }
359 
360 static int
361 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
362 		   u16 rxq_index, u32 flow_id)
363 {
364 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
365 	struct mlx4_en_filter *filter;
366 	const struct iphdr *ip;
367 	const __be16 *ports;
368 	u8 ip_proto;
369 	__be32 src_ip;
370 	__be32 dst_ip;
371 	__be16 src_port;
372 	__be16 dst_port;
373 	int nhoff = skb_network_offset(skb);
374 	int ret = 0;
375 
376 	if (skb->encapsulation)
377 		return -EPROTONOSUPPORT;
378 
379 	if (skb->protocol != htons(ETH_P_IP))
380 		return -EPROTONOSUPPORT;
381 
382 	ip = (const struct iphdr *)(skb->data + nhoff);
383 	if (ip_is_fragment(ip))
384 		return -EPROTONOSUPPORT;
385 
386 	if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
387 		return -EPROTONOSUPPORT;
388 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
389 
390 	ip_proto = ip->protocol;
391 	src_ip = ip->saddr;
392 	dst_ip = ip->daddr;
393 	src_port = ports[0];
394 	dst_port = ports[1];
395 
396 	spin_lock_bh(&priv->filters_lock);
397 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
398 				     src_port, dst_port);
399 	if (filter) {
400 		if (filter->rxq_index == rxq_index)
401 			goto out;
402 
403 		filter->rxq_index = rxq_index;
404 	} else {
405 		filter = mlx4_en_filter_alloc(priv, rxq_index,
406 					      src_ip, dst_ip, ip_proto,
407 					      src_port, dst_port, flow_id);
408 		if (!filter) {
409 			ret = -ENOMEM;
410 			goto err;
411 		}
412 	}
413 
414 	queue_work(priv->mdev->workqueue, &filter->work);
415 
416 out:
417 	ret = filter->id;
418 err:
419 	spin_unlock_bh(&priv->filters_lock);
420 
421 	return ret;
422 }
423 
424 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
425 {
426 	struct mlx4_en_filter *filter, *tmp;
427 	LIST_HEAD(del_list);
428 
429 	spin_lock_bh(&priv->filters_lock);
430 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
431 		list_move(&filter->next, &del_list);
432 		hlist_del(&filter->filter_chain);
433 	}
434 	spin_unlock_bh(&priv->filters_lock);
435 
436 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
437 		cancel_work_sync(&filter->work);
438 		mlx4_en_filter_free(filter);
439 	}
440 }
441 
442 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
443 {
444 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
445 	LIST_HEAD(del_list);
446 	int i = 0;
447 
448 	spin_lock_bh(&priv->filters_lock);
449 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
450 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
451 			break;
452 
453 		if (filter->activated &&
454 		    !work_pending(&filter->work) &&
455 		    rps_may_expire_flow(priv->dev,
456 					filter->rxq_index, filter->flow_id,
457 					filter->id)) {
458 			list_move(&filter->next, &del_list);
459 			hlist_del(&filter->filter_chain);
460 		} else
461 			last_filter = filter;
462 
463 		i++;
464 	}
465 
466 	if (last_filter && (&last_filter->next != priv->filters.next))
467 		list_move(&priv->filters, &last_filter->next);
468 
469 	spin_unlock_bh(&priv->filters_lock);
470 
471 	list_for_each_entry_safe(filter, tmp, &del_list, next)
472 		mlx4_en_filter_free(filter);
473 }
474 #endif
475 
476 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
477 				   __be16 proto, u16 vid)
478 {
479 	struct mlx4_en_priv *priv = netdev_priv(dev);
480 	struct mlx4_en_dev *mdev = priv->mdev;
481 	int err;
482 	int idx;
483 
484 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
485 
486 	set_bit(vid, priv->active_vlans);
487 
488 	/* Add VID to port VLAN filter */
489 	mutex_lock(&mdev->state_lock);
490 	if (mdev->device_up && priv->port_up) {
491 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
492 		if (err) {
493 			en_err(priv, "Failed configuring VLAN filter\n");
494 			goto out;
495 		}
496 	}
497 	err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
498 	if (err)
499 		en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
500 
501 out:
502 	mutex_unlock(&mdev->state_lock);
503 	return err;
504 }
505 
506 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
507 				    __be16 proto, u16 vid)
508 {
509 	struct mlx4_en_priv *priv = netdev_priv(dev);
510 	struct mlx4_en_dev *mdev = priv->mdev;
511 	int err = 0;
512 
513 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
514 
515 	clear_bit(vid, priv->active_vlans);
516 
517 	/* Remove VID from port VLAN filter */
518 	mutex_lock(&mdev->state_lock);
519 	mlx4_unregister_vlan(mdev->dev, priv->port, vid);
520 
521 	if (mdev->device_up && priv->port_up) {
522 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
523 		if (err)
524 			en_err(priv, "Failed configuring VLAN filter\n");
525 	}
526 	mutex_unlock(&mdev->state_lock);
527 
528 	return err;
529 }
530 
531 static void mlx4_en_u64_to_mac(struct net_device *dev, u64 src_mac)
532 {
533 	u8 addr[ETH_ALEN];
534 
535 	u64_to_ether_addr(src_mac, addr);
536 	eth_hw_addr_set(dev, addr);
537 }
538 
539 
540 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv,
541 				    const unsigned char *addr,
542 				    int qpn, u64 *reg_id)
543 {
544 	int err;
545 
546 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
547 	    priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
548 		return 0; /* do nothing */
549 
550 	err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
551 				    MLX4_DOMAIN_NIC, reg_id);
552 	if (err) {
553 		en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
554 		return err;
555 	}
556 	en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
557 	return 0;
558 }
559 
560 
561 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
562 				const unsigned char *mac, int *qpn, u64 *reg_id)
563 {
564 	struct mlx4_en_dev *mdev = priv->mdev;
565 	struct mlx4_dev *dev = mdev->dev;
566 	int err;
567 
568 	switch (dev->caps.steering_mode) {
569 	case MLX4_STEERING_MODE_B0: {
570 		struct mlx4_qp qp;
571 		u8 gid[16] = {0};
572 
573 		qp.qpn = *qpn;
574 		memcpy(&gid[10], mac, ETH_ALEN);
575 		gid[5] = priv->port;
576 
577 		err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
578 		break;
579 	}
580 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
581 		struct mlx4_spec_list spec_eth = { {NULL} };
582 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
583 
584 		struct mlx4_net_trans_rule rule = {
585 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
586 			.exclusive = 0,
587 			.allow_loopback = 1,
588 			.promisc_mode = MLX4_FS_REGULAR,
589 			.priority = MLX4_DOMAIN_NIC,
590 		};
591 
592 		rule.port = priv->port;
593 		rule.qpn = *qpn;
594 		INIT_LIST_HEAD(&rule.list);
595 
596 		spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
597 		memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
598 		memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
599 		list_add_tail(&spec_eth.list, &rule.list);
600 
601 		err = mlx4_flow_attach(dev, &rule, reg_id);
602 		break;
603 	}
604 	default:
605 		return -EINVAL;
606 	}
607 	if (err)
608 		en_warn(priv, "Failed Attaching Unicast\n");
609 
610 	return err;
611 }
612 
613 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
614 				     const unsigned char *mac,
615 				     int qpn, u64 reg_id)
616 {
617 	struct mlx4_en_dev *mdev = priv->mdev;
618 	struct mlx4_dev *dev = mdev->dev;
619 
620 	switch (dev->caps.steering_mode) {
621 	case MLX4_STEERING_MODE_B0: {
622 		struct mlx4_qp qp;
623 		u8 gid[16] = {0};
624 
625 		qp.qpn = qpn;
626 		memcpy(&gid[10], mac, ETH_ALEN);
627 		gid[5] = priv->port;
628 
629 		mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
630 		break;
631 	}
632 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
633 		mlx4_flow_detach(dev, reg_id);
634 		break;
635 	}
636 	default:
637 		en_err(priv, "Invalid steering mode.\n");
638 	}
639 }
640 
641 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
642 {
643 	struct mlx4_en_dev *mdev = priv->mdev;
644 	struct mlx4_dev *dev = mdev->dev;
645 	int index = 0;
646 	int err = 0;
647 	int *qpn = &priv->base_qpn;
648 	u64 mac = ether_addr_to_u64(priv->dev->dev_addr);
649 
650 	en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
651 	       priv->dev->dev_addr);
652 	index = mlx4_register_mac(dev, priv->port, mac);
653 	if (index < 0) {
654 		err = index;
655 		en_err(priv, "Failed adding MAC: %pM\n",
656 		       priv->dev->dev_addr);
657 		return err;
658 	}
659 
660 	en_info(priv, "Steering Mode %d\n", dev->caps.steering_mode);
661 
662 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
663 		int base_qpn = mlx4_get_base_qpn(dev, priv->port);
664 		*qpn = base_qpn + index;
665 		return 0;
666 	}
667 
668 	err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP,
669 				    MLX4_RES_USAGE_DRIVER);
670 	en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
671 	if (err) {
672 		en_err(priv, "Failed to reserve qp for mac registration\n");
673 		mlx4_unregister_mac(dev, priv->port, mac);
674 		return err;
675 	}
676 
677 	return 0;
678 }
679 
680 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
681 {
682 	struct mlx4_en_dev *mdev = priv->mdev;
683 	struct mlx4_dev *dev = mdev->dev;
684 	int qpn = priv->base_qpn;
685 
686 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
687 		u64 mac = ether_addr_to_u64(priv->dev->dev_addr);
688 		en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
689 		       priv->dev->dev_addr);
690 		mlx4_unregister_mac(dev, priv->port, mac);
691 	} else {
692 		en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
693 		       priv->port, qpn);
694 		mlx4_qp_release_range(dev, qpn, 1);
695 		priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
696 	}
697 }
698 
699 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
700 			       unsigned char *new_mac, unsigned char *prev_mac)
701 {
702 	struct mlx4_en_dev *mdev = priv->mdev;
703 	struct mlx4_dev *dev = mdev->dev;
704 	int err = 0;
705 	u64 new_mac_u64 = ether_addr_to_u64(new_mac);
706 
707 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
708 		struct hlist_head *bucket;
709 		unsigned int mac_hash;
710 		struct mlx4_mac_entry *entry;
711 		struct hlist_node *tmp;
712 		u64 prev_mac_u64 = ether_addr_to_u64(prev_mac);
713 
714 		bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
715 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
716 			if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
717 				mlx4_en_uc_steer_release(priv, entry->mac,
718 							 qpn, entry->reg_id);
719 				mlx4_unregister_mac(dev, priv->port,
720 						    prev_mac_u64);
721 				hlist_del_rcu(&entry->hlist);
722 				synchronize_rcu();
723 				memcpy(entry->mac, new_mac, ETH_ALEN);
724 				entry->reg_id = 0;
725 				mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
726 				hlist_add_head_rcu(&entry->hlist,
727 						   &priv->mac_hash[mac_hash]);
728 				mlx4_register_mac(dev, priv->port, new_mac_u64);
729 				err = mlx4_en_uc_steer_add(priv, new_mac,
730 							   &qpn,
731 							   &entry->reg_id);
732 				if (err)
733 					return err;
734 				if (priv->tunnel_reg_id) {
735 					mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
736 					priv->tunnel_reg_id = 0;
737 				}
738 				err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
739 							       &priv->tunnel_reg_id);
740 				return err;
741 			}
742 		}
743 		return -EINVAL;
744 	}
745 
746 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
747 }
748 
749 static void mlx4_en_update_user_mac(struct mlx4_en_priv *priv,
750 				    unsigned char new_mac[ETH_ALEN + 2])
751 {
752 	struct mlx4_en_dev *mdev = priv->mdev;
753 	int err;
754 
755 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_USER_MAC_EN))
756 		return;
757 
758 	err = mlx4_SET_PORT_user_mac(mdev->dev, priv->port, new_mac);
759 	if (err)
760 		en_err(priv, "Failed to pass user MAC(%pM) to Firmware for port %d, with error %d\n",
761 		       new_mac, priv->port, err);
762 }
763 
764 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
765 			      unsigned char new_mac[ETH_ALEN + 2])
766 {
767 	int err = 0;
768 
769 	if (priv->port_up) {
770 		/* Remove old MAC and insert the new one */
771 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
772 					  new_mac, priv->current_mac);
773 		if (err)
774 			en_err(priv, "Failed changing HW MAC address\n");
775 	} else
776 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
777 
778 	if (!err)
779 		memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
780 
781 	return err;
782 }
783 
784 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
785 {
786 	struct mlx4_en_priv *priv = netdev_priv(dev);
787 	struct mlx4_en_dev *mdev = priv->mdev;
788 	struct sockaddr *saddr = addr;
789 	unsigned char new_mac[ETH_ALEN + 2];
790 	int err;
791 
792 	if (!is_valid_ether_addr(saddr->sa_data))
793 		return -EADDRNOTAVAIL;
794 
795 	mutex_lock(&mdev->state_lock);
796 	memcpy(new_mac, saddr->sa_data, ETH_ALEN);
797 	err = mlx4_en_do_set_mac(priv, new_mac);
798 	if (err)
799 		goto out;
800 
801 	eth_hw_addr_set(dev, saddr->sa_data);
802 	mlx4_en_update_user_mac(priv, new_mac);
803 out:
804 	mutex_unlock(&mdev->state_lock);
805 
806 	return err;
807 }
808 
809 static void mlx4_en_clear_list(struct net_device *dev)
810 {
811 	struct mlx4_en_priv *priv = netdev_priv(dev);
812 	struct mlx4_en_mc_list *tmp, *mc_to_del;
813 
814 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
815 		list_del(&mc_to_del->list);
816 		kfree(mc_to_del);
817 	}
818 }
819 
820 static void mlx4_en_cache_mclist(struct net_device *dev)
821 {
822 	struct mlx4_en_priv *priv = netdev_priv(dev);
823 	struct netdev_hw_addr *ha;
824 	struct mlx4_en_mc_list *tmp;
825 
826 	mlx4_en_clear_list(dev);
827 	netdev_for_each_mc_addr(ha, dev) {
828 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
829 		if (!tmp) {
830 			mlx4_en_clear_list(dev);
831 			return;
832 		}
833 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
834 		list_add_tail(&tmp->list, &priv->mc_list);
835 	}
836 }
837 
838 static void update_mclist_flags(struct mlx4_en_priv *priv,
839 				struct list_head *dst,
840 				struct list_head *src)
841 {
842 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
843 	bool found;
844 
845 	/* Find all the entries that should be removed from dst,
846 	 * These are the entries that are not found in src
847 	 */
848 	list_for_each_entry(dst_tmp, dst, list) {
849 		found = false;
850 		list_for_each_entry(src_tmp, src, list) {
851 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
852 				found = true;
853 				break;
854 			}
855 		}
856 		if (!found)
857 			dst_tmp->action = MCLIST_REM;
858 	}
859 
860 	/* Add entries that exist in src but not in dst
861 	 * mark them as need to add
862 	 */
863 	list_for_each_entry(src_tmp, src, list) {
864 		found = false;
865 		list_for_each_entry(dst_tmp, dst, list) {
866 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
867 				dst_tmp->action = MCLIST_NONE;
868 				found = true;
869 				break;
870 			}
871 		}
872 		if (!found) {
873 			new_mc = kmemdup(src_tmp,
874 					 sizeof(struct mlx4_en_mc_list),
875 					 GFP_KERNEL);
876 			if (!new_mc)
877 				return;
878 
879 			new_mc->action = MCLIST_ADD;
880 			list_add_tail(&new_mc->list, dst);
881 		}
882 	}
883 }
884 
885 static void mlx4_en_set_rx_mode(struct net_device *dev)
886 {
887 	struct mlx4_en_priv *priv = netdev_priv(dev);
888 
889 	if (!priv->port_up)
890 		return;
891 
892 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
893 }
894 
895 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
896 				     struct mlx4_en_dev *mdev)
897 {
898 	int err = 0;
899 
900 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
901 		if (netif_msg_rx_status(priv))
902 			en_warn(priv, "Entering promiscuous mode\n");
903 		priv->flags |= MLX4_EN_FLAG_PROMISC;
904 
905 		/* Enable promiscouos mode */
906 		switch (mdev->dev->caps.steering_mode) {
907 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
908 			err = mlx4_flow_steer_promisc_add(mdev->dev,
909 							  priv->port,
910 							  priv->base_qpn,
911 							  MLX4_FS_ALL_DEFAULT);
912 			if (err)
913 				en_err(priv, "Failed enabling promiscuous mode\n");
914 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
915 			break;
916 
917 		case MLX4_STEERING_MODE_B0:
918 			err = mlx4_unicast_promisc_add(mdev->dev,
919 						       priv->base_qpn,
920 						       priv->port);
921 			if (err)
922 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
923 
924 			/* Add the default qp number as multicast
925 			 * promisc
926 			 */
927 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
928 				err = mlx4_multicast_promisc_add(mdev->dev,
929 								 priv->base_qpn,
930 								 priv->port);
931 				if (err)
932 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
933 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
934 			}
935 			break;
936 
937 		case MLX4_STEERING_MODE_A0:
938 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
939 						     priv->port,
940 						     priv->base_qpn,
941 						     1);
942 			if (err)
943 				en_err(priv, "Failed enabling promiscuous mode\n");
944 			break;
945 		}
946 
947 		/* Disable port multicast filter (unconditionally) */
948 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
949 					  0, MLX4_MCAST_DISABLE);
950 		if (err)
951 			en_err(priv, "Failed disabling multicast filter\n");
952 	}
953 }
954 
955 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
956 				       struct mlx4_en_dev *mdev)
957 {
958 	int err = 0;
959 
960 	if (netif_msg_rx_status(priv))
961 		en_warn(priv, "Leaving promiscuous mode\n");
962 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
963 
964 	/* Disable promiscouos mode */
965 	switch (mdev->dev->caps.steering_mode) {
966 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
967 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
968 						     priv->port,
969 						     MLX4_FS_ALL_DEFAULT);
970 		if (err)
971 			en_err(priv, "Failed disabling promiscuous mode\n");
972 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
973 		break;
974 
975 	case MLX4_STEERING_MODE_B0:
976 		err = mlx4_unicast_promisc_remove(mdev->dev,
977 						  priv->base_qpn,
978 						  priv->port);
979 		if (err)
980 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
981 		/* Disable Multicast promisc */
982 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
983 			err = mlx4_multicast_promisc_remove(mdev->dev,
984 							    priv->base_qpn,
985 							    priv->port);
986 			if (err)
987 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
988 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
989 		}
990 		break;
991 
992 	case MLX4_STEERING_MODE_A0:
993 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
994 					     priv->port,
995 					     priv->base_qpn, 0);
996 		if (err)
997 			en_err(priv, "Failed disabling promiscuous mode\n");
998 		break;
999 	}
1000 }
1001 
1002 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
1003 				 struct net_device *dev,
1004 				 struct mlx4_en_dev *mdev)
1005 {
1006 	struct mlx4_en_mc_list *mclist, *tmp;
1007 	u64 mcast_addr = 0;
1008 	u8 mc_list[16] = {0};
1009 	int err = 0;
1010 
1011 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
1012 	if (dev->flags & IFF_ALLMULTI) {
1013 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1014 					  0, MLX4_MCAST_DISABLE);
1015 		if (err)
1016 			en_err(priv, "Failed disabling multicast filter\n");
1017 
1018 		/* Add the default qp number as multicast promisc */
1019 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1020 			switch (mdev->dev->caps.steering_mode) {
1021 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1022 				err = mlx4_flow_steer_promisc_add(mdev->dev,
1023 								  priv->port,
1024 								  priv->base_qpn,
1025 								  MLX4_FS_MC_DEFAULT);
1026 				break;
1027 
1028 			case MLX4_STEERING_MODE_B0:
1029 				err = mlx4_multicast_promisc_add(mdev->dev,
1030 								 priv->base_qpn,
1031 								 priv->port);
1032 				break;
1033 
1034 			case MLX4_STEERING_MODE_A0:
1035 				break;
1036 			}
1037 			if (err)
1038 				en_err(priv, "Failed entering multicast promisc mode\n");
1039 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1040 		}
1041 	} else {
1042 		/* Disable Multicast promisc */
1043 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1044 			switch (mdev->dev->caps.steering_mode) {
1045 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1046 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
1047 								     priv->port,
1048 								     MLX4_FS_MC_DEFAULT);
1049 				break;
1050 
1051 			case MLX4_STEERING_MODE_B0:
1052 				err = mlx4_multicast_promisc_remove(mdev->dev,
1053 								    priv->base_qpn,
1054 								    priv->port);
1055 				break;
1056 
1057 			case MLX4_STEERING_MODE_A0:
1058 				break;
1059 			}
1060 			if (err)
1061 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
1062 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1063 		}
1064 
1065 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1066 					  0, MLX4_MCAST_DISABLE);
1067 		if (err)
1068 			en_err(priv, "Failed disabling multicast filter\n");
1069 
1070 		/* Flush mcast filter and init it with broadcast address */
1071 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1072 				    1, MLX4_MCAST_CONFIG);
1073 
1074 		/* Update multicast list - we cache all addresses so they won't
1075 		 * change while HW is updated holding the command semaphor */
1076 		netif_addr_lock_bh(dev);
1077 		mlx4_en_cache_mclist(dev);
1078 		netif_addr_unlock_bh(dev);
1079 		list_for_each_entry(mclist, &priv->mc_list, list) {
1080 			mcast_addr = ether_addr_to_u64(mclist->addr);
1081 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1082 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
1083 		}
1084 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1085 					  0, MLX4_MCAST_ENABLE);
1086 		if (err)
1087 			en_err(priv, "Failed enabling multicast filter\n");
1088 
1089 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1090 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1091 			if (mclist->action == MCLIST_REM) {
1092 				/* detach this address and delete from list */
1093 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1094 				mc_list[5] = priv->port;
1095 				err = mlx4_multicast_detach(mdev->dev,
1096 							    priv->rss_map.indir_qp,
1097 							    mc_list,
1098 							    MLX4_PROT_ETH,
1099 							    mclist->reg_id);
1100 				if (err)
1101 					en_err(priv, "Fail to detach multicast address\n");
1102 
1103 				if (mclist->tunnel_reg_id) {
1104 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1105 					if (err)
1106 						en_err(priv, "Failed to detach multicast address\n");
1107 				}
1108 
1109 				/* remove from list */
1110 				list_del(&mclist->list);
1111 				kfree(mclist);
1112 			} else if (mclist->action == MCLIST_ADD) {
1113 				/* attach the address */
1114 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1115 				/* needed for B0 steering support */
1116 				mc_list[5] = priv->port;
1117 				err = mlx4_multicast_attach(mdev->dev,
1118 							    priv->rss_map.indir_qp,
1119 							    mc_list,
1120 							    priv->port, 0,
1121 							    MLX4_PROT_ETH,
1122 							    &mclist->reg_id);
1123 				if (err)
1124 					en_err(priv, "Fail to attach multicast address\n");
1125 
1126 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1127 							       &mclist->tunnel_reg_id);
1128 				if (err)
1129 					en_err(priv, "Failed to attach multicast address\n");
1130 			}
1131 		}
1132 	}
1133 }
1134 
1135 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1136 				 struct net_device *dev,
1137 				 struct mlx4_en_dev *mdev)
1138 {
1139 	struct netdev_hw_addr *ha;
1140 	struct mlx4_mac_entry *entry;
1141 	struct hlist_node *tmp;
1142 	bool found;
1143 	u64 mac;
1144 	int err = 0;
1145 	struct hlist_head *bucket;
1146 	unsigned int i;
1147 	int removed = 0;
1148 	u32 prev_flags;
1149 
1150 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1151 	 * since all modification code is protected by mdev->state_lock
1152 	 */
1153 
1154 	/* find what to remove */
1155 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1156 		bucket = &priv->mac_hash[i];
1157 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1158 			found = false;
1159 			netdev_for_each_uc_addr(ha, dev) {
1160 				if (ether_addr_equal_64bits(entry->mac,
1161 							    ha->addr)) {
1162 					found = true;
1163 					break;
1164 				}
1165 			}
1166 
1167 			/* MAC address of the port is not in uc list */
1168 			if (ether_addr_equal_64bits(entry->mac,
1169 						    priv->current_mac))
1170 				found = true;
1171 
1172 			if (!found) {
1173 				mac = ether_addr_to_u64(entry->mac);
1174 				mlx4_en_uc_steer_release(priv, entry->mac,
1175 							 priv->base_qpn,
1176 							 entry->reg_id);
1177 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1178 
1179 				hlist_del_rcu(&entry->hlist);
1180 				kfree_rcu(entry, rcu);
1181 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1182 				       entry->mac, priv->port);
1183 				++removed;
1184 			}
1185 		}
1186 	}
1187 
1188 	/* if we didn't remove anything, there is no use in trying to add
1189 	 * again once we are in a forced promisc mode state
1190 	 */
1191 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1192 		return;
1193 
1194 	prev_flags = priv->flags;
1195 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1196 
1197 	/* find what to add */
1198 	netdev_for_each_uc_addr(ha, dev) {
1199 		found = false;
1200 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1201 		hlist_for_each_entry(entry, bucket, hlist) {
1202 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1203 				found = true;
1204 				break;
1205 			}
1206 		}
1207 
1208 		if (!found) {
1209 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1210 			if (!entry) {
1211 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1212 				       ha->addr, priv->port);
1213 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1214 				break;
1215 			}
1216 			mac = ether_addr_to_u64(ha->addr);
1217 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1218 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1219 			if (err < 0) {
1220 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1221 				       ha->addr, priv->port, err);
1222 				kfree(entry);
1223 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1224 				break;
1225 			}
1226 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1227 						   &priv->base_qpn,
1228 						   &entry->reg_id);
1229 			if (err) {
1230 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1231 				       ha->addr, priv->port, err);
1232 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1233 				kfree(entry);
1234 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1235 				break;
1236 			} else {
1237 				unsigned int mac_hash;
1238 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1239 				       ha->addr, priv->port);
1240 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1241 				bucket = &priv->mac_hash[mac_hash];
1242 				hlist_add_head_rcu(&entry->hlist, bucket);
1243 			}
1244 		}
1245 	}
1246 
1247 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1248 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1249 			priv->port);
1250 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1251 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1252 			priv->port);
1253 	}
1254 }
1255 
1256 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1257 {
1258 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1259 						 rx_mode_task);
1260 	struct mlx4_en_dev *mdev = priv->mdev;
1261 	struct net_device *dev = priv->dev;
1262 
1263 	mutex_lock(&mdev->state_lock);
1264 	if (!mdev->device_up) {
1265 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1266 		goto out;
1267 	}
1268 	if (!priv->port_up) {
1269 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1270 		goto out;
1271 	}
1272 
1273 	if (!netif_carrier_ok(dev)) {
1274 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1275 			if (priv->port_state.link_state) {
1276 				netif_carrier_on(dev);
1277 				en_dbg(LINK, priv, "Link Up\n");
1278 			}
1279 		}
1280 	}
1281 
1282 	if (dev->priv_flags & IFF_UNICAST_FLT)
1283 		mlx4_en_do_uc_filter(priv, dev, mdev);
1284 
1285 	/* Promsicuous mode: disable all filters */
1286 	if ((dev->flags & IFF_PROMISC) ||
1287 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1288 		mlx4_en_set_promisc_mode(priv, mdev);
1289 		goto out;
1290 	}
1291 
1292 	/* Not in promiscuous mode */
1293 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1294 		mlx4_en_clear_promisc_mode(priv, mdev);
1295 
1296 	mlx4_en_do_multicast(priv, dev, mdev);
1297 out:
1298 	mutex_unlock(&mdev->state_lock);
1299 }
1300 
1301 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1302 {
1303 	u64 reg_id;
1304 	int err = 0;
1305 	int *qpn = &priv->base_qpn;
1306 	struct mlx4_mac_entry *entry;
1307 
1308 	err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1309 	if (err)
1310 		return err;
1311 
1312 	err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1313 				       &priv->tunnel_reg_id);
1314 	if (err)
1315 		goto tunnel_err;
1316 
1317 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1318 	if (!entry) {
1319 		err = -ENOMEM;
1320 		goto alloc_err;
1321 	}
1322 
1323 	memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1324 	memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1325 	entry->reg_id = reg_id;
1326 	hlist_add_head_rcu(&entry->hlist,
1327 			   &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1328 
1329 	return 0;
1330 
1331 alloc_err:
1332 	if (priv->tunnel_reg_id)
1333 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1334 
1335 tunnel_err:
1336 	mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1337 	return err;
1338 }
1339 
1340 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1341 {
1342 	u64 mac;
1343 	unsigned int i;
1344 	int qpn = priv->base_qpn;
1345 	struct hlist_head *bucket;
1346 	struct hlist_node *tmp;
1347 	struct mlx4_mac_entry *entry;
1348 
1349 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1350 		bucket = &priv->mac_hash[i];
1351 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1352 			mac = ether_addr_to_u64(entry->mac);
1353 			en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1354 			       entry->mac);
1355 			mlx4_en_uc_steer_release(priv, entry->mac,
1356 						 qpn, entry->reg_id);
1357 
1358 			mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1359 			hlist_del_rcu(&entry->hlist);
1360 			kfree_rcu(entry, rcu);
1361 		}
1362 	}
1363 
1364 	if (priv->tunnel_reg_id) {
1365 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1366 		priv->tunnel_reg_id = 0;
1367 	}
1368 }
1369 
1370 static void mlx4_en_tx_timeout(struct net_device *dev, unsigned int txqueue)
1371 {
1372 	struct mlx4_en_priv *priv = netdev_priv(dev);
1373 	struct mlx4_en_dev *mdev = priv->mdev;
1374 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][txqueue];
1375 
1376 	if (netif_msg_timer(priv))
1377 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1378 
1379 	en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1380 		txqueue, tx_ring->qpn, tx_ring->sp_cqn,
1381 		tx_ring->cons, tx_ring->prod);
1382 
1383 	priv->port_stats.tx_timeout++;
1384 	if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state)) {
1385 		en_dbg(DRV, priv, "Scheduling port restart\n");
1386 		queue_work(mdev->workqueue, &priv->restart_task);
1387 	}
1388 }
1389 
1390 
1391 static void
1392 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1393 {
1394 	struct mlx4_en_priv *priv = netdev_priv(dev);
1395 
1396 	spin_lock_bh(&priv->stats_lock);
1397 	mlx4_en_fold_software_stats(dev);
1398 	netdev_stats_to_stats64(stats, &dev->stats);
1399 	spin_unlock_bh(&priv->stats_lock);
1400 }
1401 
1402 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1403 {
1404 	struct mlx4_en_cq *cq;
1405 	int i, t;
1406 
1407 	/* If we haven't received a specific coalescing setting
1408 	 * (module param), we set the moderation parameters as follows:
1409 	 * - moder_cnt is set to the number of mtu sized packets to
1410 	 *   satisfy our coalescing target.
1411 	 * - moder_time is set to a fixed value.
1412 	 */
1413 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1414 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1415 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1416 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1417 	en_dbg(INTR, priv, "Default coalescing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1418 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1419 
1420 	/* Setup cq moderation params */
1421 	for (i = 0; i < priv->rx_ring_num; i++) {
1422 		cq = priv->rx_cq[i];
1423 		cq->moder_cnt = priv->rx_frames;
1424 		cq->moder_time = priv->rx_usecs;
1425 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1426 		priv->last_moder_packets[i] = 0;
1427 		priv->last_moder_bytes[i] = 0;
1428 	}
1429 
1430 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1431 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1432 			cq = priv->tx_cq[t][i];
1433 			cq->moder_cnt = priv->tx_frames;
1434 			cq->moder_time = priv->tx_usecs;
1435 		}
1436 	}
1437 
1438 	/* Reset auto-moderation params */
1439 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1440 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1441 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1442 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1443 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1444 	priv->adaptive_rx_coal = 1;
1445 	priv->last_moder_jiffies = 0;
1446 	priv->last_moder_tx_packets = 0;
1447 }
1448 
1449 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1450 {
1451 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1452 	u32 pkt_rate_high, pkt_rate_low;
1453 	struct mlx4_en_cq *cq;
1454 	unsigned long packets;
1455 	unsigned long rate;
1456 	unsigned long avg_pkt_size;
1457 	unsigned long rx_packets;
1458 	unsigned long rx_bytes;
1459 	unsigned long rx_pkt_diff;
1460 	int moder_time;
1461 	int ring, err;
1462 
1463 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1464 		return;
1465 
1466 	pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1467 	pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1468 
1469 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1470 		rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1471 		rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1472 
1473 		rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1474 		packets = rx_pkt_diff;
1475 		rate = packets * HZ / period;
1476 		avg_pkt_size = packets ? (rx_bytes -
1477 				priv->last_moder_bytes[ring]) / packets : 0;
1478 
1479 		/* Apply auto-moderation only when packet rate
1480 		 * exceeds a rate that it matters */
1481 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1482 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1483 			if (rate <= pkt_rate_low)
1484 				moder_time = priv->rx_usecs_low;
1485 			else if (rate >= pkt_rate_high)
1486 				moder_time = priv->rx_usecs_high;
1487 			else
1488 				moder_time = (rate - pkt_rate_low) *
1489 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1490 					(pkt_rate_high - pkt_rate_low) +
1491 					priv->rx_usecs_low;
1492 		} else {
1493 			moder_time = priv->rx_usecs_low;
1494 		}
1495 
1496 		cq = priv->rx_cq[ring];
1497 		if (moder_time != priv->last_moder_time[ring] ||
1498 		    cq->moder_cnt != priv->rx_frames) {
1499 			priv->last_moder_time[ring] = moder_time;
1500 			cq->moder_time = moder_time;
1501 			cq->moder_cnt = priv->rx_frames;
1502 			err = mlx4_en_set_cq_moder(priv, cq);
1503 			if (err)
1504 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1505 				       ring);
1506 		}
1507 		priv->last_moder_packets[ring] = rx_packets;
1508 		priv->last_moder_bytes[ring] = rx_bytes;
1509 	}
1510 
1511 	priv->last_moder_jiffies = jiffies;
1512 }
1513 
1514 static void mlx4_en_do_get_stats(struct work_struct *work)
1515 {
1516 	struct delayed_work *delay = to_delayed_work(work);
1517 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1518 						 stats_task);
1519 	struct mlx4_en_dev *mdev = priv->mdev;
1520 	int err;
1521 
1522 	mutex_lock(&mdev->state_lock);
1523 	if (mdev->device_up) {
1524 		if (priv->port_up) {
1525 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1526 			if (err)
1527 				en_dbg(HW, priv, "Could not update stats\n");
1528 
1529 			mlx4_en_auto_moderation(priv);
1530 		}
1531 
1532 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1533 	}
1534 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1535 		mlx4_en_do_set_mac(priv, priv->current_mac);
1536 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1537 	}
1538 	mutex_unlock(&mdev->state_lock);
1539 }
1540 
1541 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1542  * periodically
1543  */
1544 static void mlx4_en_service_task(struct work_struct *work)
1545 {
1546 	struct delayed_work *delay = to_delayed_work(work);
1547 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1548 						 service_task);
1549 	struct mlx4_en_dev *mdev = priv->mdev;
1550 
1551 	mutex_lock(&mdev->state_lock);
1552 	if (mdev->device_up) {
1553 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1554 			mlx4_en_ptp_overflow_check(mdev);
1555 
1556 		mlx4_en_recover_from_oom(priv);
1557 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1558 				   SERVICE_TASK_DELAY);
1559 	}
1560 	mutex_unlock(&mdev->state_lock);
1561 }
1562 
1563 static void mlx4_en_linkstate(struct mlx4_en_priv *priv)
1564 {
1565 	struct mlx4_en_port_state *port_state = &priv->port_state;
1566 	struct mlx4_en_dev *mdev = priv->mdev;
1567 	struct net_device *dev = priv->dev;
1568 	bool up;
1569 
1570 	if (mlx4_en_QUERY_PORT(mdev, priv->port))
1571 		port_state->link_state = MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN;
1572 
1573 	up = port_state->link_state == MLX4_PORT_STATE_DEV_EVENT_PORT_UP;
1574 	if (up == netif_carrier_ok(dev))
1575 		netif_carrier_event(dev);
1576 	if (!up) {
1577 		en_info(priv, "Link Down\n");
1578 		netif_carrier_off(dev);
1579 	} else {
1580 		en_info(priv, "Link Up\n");
1581 		netif_carrier_on(dev);
1582 	}
1583 }
1584 
1585 static void mlx4_en_linkstate_work(struct work_struct *work)
1586 {
1587 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1588 						 linkstate_task);
1589 	struct mlx4_en_dev *mdev = priv->mdev;
1590 
1591 	mutex_lock(&mdev->state_lock);
1592 	mlx4_en_linkstate(priv);
1593 	mutex_unlock(&mdev->state_lock);
1594 }
1595 
1596 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1597 {
1598 	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1599 	int numa_node = priv->mdev->dev->numa_node;
1600 
1601 	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1602 		return -ENOMEM;
1603 
1604 	cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1605 			ring->affinity_mask);
1606 	return 0;
1607 }
1608 
1609 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1610 {
1611 	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1612 }
1613 
1614 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1615 				      int tx_ring_idx)
1616 {
1617 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1618 	int rr_index = tx_ring_idx;
1619 
1620 	tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1621 	tx_ring->recycle_ring = priv->rx_ring[rr_index];
1622 	en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1623 	       TX_XDP, tx_ring_idx, rr_index);
1624 }
1625 
1626 int mlx4_en_start_port(struct net_device *dev)
1627 {
1628 	struct mlx4_en_priv *priv = netdev_priv(dev);
1629 	struct mlx4_en_dev *mdev = priv->mdev;
1630 	struct mlx4_en_cq *cq;
1631 	struct mlx4_en_tx_ring *tx_ring;
1632 	int rx_index = 0;
1633 	int err = 0;
1634 	int i, t;
1635 	int j;
1636 	u8 mc_list[16] = {0};
1637 
1638 	if (priv->port_up) {
1639 		en_dbg(DRV, priv, "start port called while port already up\n");
1640 		return 0;
1641 	}
1642 
1643 	INIT_LIST_HEAD(&priv->mc_list);
1644 	INIT_LIST_HEAD(&priv->curr_list);
1645 	INIT_LIST_HEAD(&priv->ethtool_list);
1646 	memset(&priv->ethtool_rules[0], 0,
1647 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1648 
1649 	/* Calculate Rx buf size */
1650 	dev->mtu = min(dev->mtu, priv->max_mtu);
1651 	mlx4_en_calc_rx_buf(dev);
1652 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1653 
1654 	/* Configure rx cq's and rings */
1655 	err = mlx4_en_activate_rx_rings(priv);
1656 	if (err) {
1657 		en_err(priv, "Failed to activate RX rings\n");
1658 		return err;
1659 	}
1660 	for (i = 0; i < priv->rx_ring_num; i++) {
1661 		cq = priv->rx_cq[i];
1662 
1663 		err = mlx4_en_init_affinity_hint(priv, i);
1664 		if (err) {
1665 			en_err(priv, "Failed preparing IRQ affinity hint\n");
1666 			goto cq_err;
1667 		}
1668 
1669 		err = mlx4_en_activate_cq(priv, cq, i);
1670 		if (err) {
1671 			en_err(priv, "Failed activating Rx CQ\n");
1672 			mlx4_en_free_affinity_hint(priv, i);
1673 			goto cq_err;
1674 		}
1675 
1676 		for (j = 0; j < cq->size; j++) {
1677 			struct mlx4_cqe *cqe = NULL;
1678 
1679 			cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1680 			      priv->cqe_factor;
1681 			cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1682 		}
1683 
1684 		err = mlx4_en_set_cq_moder(priv, cq);
1685 		if (err) {
1686 			en_err(priv, "Failed setting cq moderation parameters\n");
1687 			mlx4_en_deactivate_cq(priv, cq);
1688 			mlx4_en_free_affinity_hint(priv, i);
1689 			goto cq_err;
1690 		}
1691 		mlx4_en_arm_cq(priv, cq);
1692 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1693 		++rx_index;
1694 	}
1695 
1696 	/* Set qp number */
1697 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1698 	err = mlx4_en_get_qp(priv);
1699 	if (err) {
1700 		en_err(priv, "Failed getting eth qp\n");
1701 		goto cq_err;
1702 	}
1703 	mdev->mac_removed[priv->port] = 0;
1704 
1705 	priv->counter_index =
1706 			mlx4_get_default_counter_index(mdev->dev, priv->port);
1707 
1708 	err = mlx4_en_config_rss_steer(priv);
1709 	if (err) {
1710 		en_err(priv, "Failed configuring rss steering\n");
1711 		goto mac_err;
1712 	}
1713 
1714 	err = mlx4_en_create_drop_qp(priv);
1715 	if (err)
1716 		goto rss_err;
1717 
1718 	/* Configure tx cq's and rings */
1719 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1720 		u8 num_tx_rings_p_up = t == TX ?
1721 			priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1722 
1723 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1724 			/* Configure cq */
1725 			cq = priv->tx_cq[t][i];
1726 			err = mlx4_en_activate_cq(priv, cq, i);
1727 			if (err) {
1728 				en_err(priv, "Failed allocating Tx CQ\n");
1729 				goto tx_err;
1730 			}
1731 			err = mlx4_en_set_cq_moder(priv, cq);
1732 			if (err) {
1733 				en_err(priv, "Failed setting cq moderation parameters\n");
1734 				mlx4_en_deactivate_cq(priv, cq);
1735 				goto tx_err;
1736 			}
1737 			en_dbg(DRV, priv,
1738 			       "Resetting index of collapsed CQ:%d to -1\n", i);
1739 			cq->buf->wqe_index = cpu_to_be16(0xffff);
1740 
1741 			/* Configure ring */
1742 			tx_ring = priv->tx_ring[t][i];
1743 			err = mlx4_en_activate_tx_ring(priv, tx_ring,
1744 						       cq->mcq.cqn,
1745 						       i / num_tx_rings_p_up);
1746 			if (err) {
1747 				en_err(priv, "Failed allocating Tx ring\n");
1748 				mlx4_en_deactivate_cq(priv, cq);
1749 				goto tx_err;
1750 			}
1751 			clear_bit(MLX4_EN_TX_RING_STATE_RECOVERING, &tx_ring->state);
1752 			if (t != TX_XDP) {
1753 				tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1754 				tx_ring->recycle_ring = NULL;
1755 
1756 				/* Arm CQ for TX completions */
1757 				mlx4_en_arm_cq(priv, cq);
1758 
1759 			} else {
1760 				mlx4_en_init_tx_xdp_ring_descs(priv, tx_ring);
1761 				mlx4_en_init_recycle_ring(priv, i);
1762 				/* XDP TX CQ should never be armed */
1763 			}
1764 
1765 			/* Set initial ownership of all Tx TXBBs to SW (1) */
1766 			for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1767 				*((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1768 		}
1769 	}
1770 
1771 	/* Configure port */
1772 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1773 				    priv->rx_skb_size + ETH_FCS_LEN,
1774 				    priv->prof->tx_pause,
1775 				    priv->prof->tx_ppp,
1776 				    priv->prof->rx_pause,
1777 				    priv->prof->rx_ppp);
1778 	if (err) {
1779 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1780 		       priv->port, err);
1781 		goto tx_err;
1782 	}
1783 
1784 	err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1785 	if (err) {
1786 		en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1787 		       dev->mtu, priv->port, err);
1788 		goto tx_err;
1789 	}
1790 
1791 	/* Set default qp number */
1792 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1793 	if (err) {
1794 		en_err(priv, "Failed setting default qp numbers\n");
1795 		goto tx_err;
1796 	}
1797 
1798 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1799 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1800 		if (err) {
1801 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1802 			       err);
1803 			goto tx_err;
1804 		}
1805 	}
1806 
1807 	/* Init port */
1808 	en_dbg(HW, priv, "Initializing port\n");
1809 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1810 	if (err) {
1811 		en_err(priv, "Failed Initializing port\n");
1812 		goto tx_err;
1813 	}
1814 
1815 	/* Set Unicast and VXLAN steering rules */
1816 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1817 	    mlx4_en_set_rss_steer_rules(priv))
1818 		mlx4_warn(mdev, "Failed setting steering rules\n");
1819 
1820 	/* Attach rx QP to bradcast address */
1821 	eth_broadcast_addr(&mc_list[10]);
1822 	mc_list[5] = priv->port; /* needed for B0 steering support */
1823 	if (mlx4_multicast_attach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1824 				  priv->port, 0, MLX4_PROT_ETH,
1825 				  &priv->broadcast_id))
1826 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1827 
1828 	/* Must redo promiscuous mode setup. */
1829 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1830 
1831 	/* Schedule multicast task to populate multicast list */
1832 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1833 
1834 	if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1835 		udp_tunnel_nic_reset_ntf(dev);
1836 
1837 	priv->port_up = true;
1838 
1839 	/* Process all completions if exist to prevent
1840 	 * the queues freezing if they are full
1841 	 */
1842 	for (i = 0; i < priv->rx_ring_num; i++) {
1843 		local_bh_disable();
1844 		napi_schedule(&priv->rx_cq[i]->napi);
1845 		local_bh_enable();
1846 	}
1847 
1848 	clear_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state);
1849 	netif_tx_start_all_queues(dev);
1850 	netif_device_attach(dev);
1851 
1852 	return 0;
1853 
1854 tx_err:
1855 	if (t == MLX4_EN_NUM_TX_TYPES) {
1856 		t--;
1857 		i = priv->tx_ring_num[t];
1858 	}
1859 	while (t >= 0) {
1860 		while (i--) {
1861 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1862 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1863 		}
1864 		if (!t--)
1865 			break;
1866 		i = priv->tx_ring_num[t];
1867 	}
1868 	mlx4_en_destroy_drop_qp(priv);
1869 rss_err:
1870 	mlx4_en_release_rss_steer(priv);
1871 mac_err:
1872 	mlx4_en_put_qp(priv);
1873 cq_err:
1874 	while (rx_index--) {
1875 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1876 		mlx4_en_free_affinity_hint(priv, rx_index);
1877 	}
1878 	for (i = 0; i < priv->rx_ring_num; i++)
1879 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1880 
1881 	return err; /* need to close devices */
1882 }
1883 
1884 
1885 void mlx4_en_stop_port(struct net_device *dev, int detach)
1886 {
1887 	struct mlx4_en_priv *priv = netdev_priv(dev);
1888 	struct mlx4_en_dev *mdev = priv->mdev;
1889 	struct mlx4_en_mc_list *mclist, *tmp;
1890 	struct ethtool_flow_id *flow, *tmp_flow;
1891 	int i, t;
1892 	u8 mc_list[16] = {0};
1893 
1894 	if (!priv->port_up) {
1895 		en_dbg(DRV, priv, "stop port called while port already down\n");
1896 		return;
1897 	}
1898 
1899 	/* close port*/
1900 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1901 
1902 	/* Synchronize with tx routine */
1903 	netif_tx_lock_bh(dev);
1904 	if (detach)
1905 		netif_device_detach(dev);
1906 	netif_tx_stop_all_queues(dev);
1907 	netif_tx_unlock_bh(dev);
1908 
1909 	netif_tx_disable(dev);
1910 
1911 	spin_lock_bh(&priv->stats_lock);
1912 	mlx4_en_fold_software_stats(dev);
1913 	/* Set port as not active */
1914 	priv->port_up = false;
1915 	spin_unlock_bh(&priv->stats_lock);
1916 
1917 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1918 
1919 	/* Promsicuous mode */
1920 	if (mdev->dev->caps.steering_mode ==
1921 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1922 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1923 				 MLX4_EN_FLAG_MC_PROMISC);
1924 		mlx4_flow_steer_promisc_remove(mdev->dev,
1925 					       priv->port,
1926 					       MLX4_FS_ALL_DEFAULT);
1927 		mlx4_flow_steer_promisc_remove(mdev->dev,
1928 					       priv->port,
1929 					       MLX4_FS_MC_DEFAULT);
1930 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1931 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1932 
1933 		/* Disable promiscouos mode */
1934 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1935 					    priv->port);
1936 
1937 		/* Disable Multicast promisc */
1938 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1939 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1940 						      priv->port);
1941 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1942 		}
1943 	}
1944 
1945 	/* Detach All multicasts */
1946 	eth_broadcast_addr(&mc_list[10]);
1947 	mc_list[5] = priv->port; /* needed for B0 steering support */
1948 	mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1949 			      MLX4_PROT_ETH, priv->broadcast_id);
1950 	list_for_each_entry(mclist, &priv->curr_list, list) {
1951 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1952 		mc_list[5] = priv->port;
1953 		mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp,
1954 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1955 		if (mclist->tunnel_reg_id)
1956 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1957 	}
1958 	mlx4_en_clear_list(dev);
1959 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1960 		list_del(&mclist->list);
1961 		kfree(mclist);
1962 	}
1963 
1964 	/* Flush multicast filter */
1965 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1966 
1967 	/* Remove flow steering rules for the port*/
1968 	if (mdev->dev->caps.steering_mode ==
1969 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1970 		ASSERT_RTNL();
1971 		list_for_each_entry_safe(flow, tmp_flow,
1972 					 &priv->ethtool_list, list) {
1973 			mlx4_flow_detach(mdev->dev, flow->id);
1974 			list_del(&flow->list);
1975 		}
1976 	}
1977 
1978 	mlx4_en_destroy_drop_qp(priv);
1979 
1980 	/* Free TX Rings */
1981 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1982 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1983 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1984 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1985 		}
1986 	}
1987 	msleep(10);
1988 
1989 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1990 		for (i = 0; i < priv->tx_ring_num[t]; i++)
1991 			mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1992 
1993 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1994 		mlx4_en_delete_rss_steer_rules(priv);
1995 
1996 	/* Free RSS qps */
1997 	mlx4_en_release_rss_steer(priv);
1998 
1999 	/* Unregister Mac address for the port */
2000 	mlx4_en_put_qp(priv);
2001 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
2002 		mdev->mac_removed[priv->port] = 1;
2003 
2004 	/* Free RX Rings */
2005 	for (i = 0; i < priv->rx_ring_num; i++) {
2006 		struct mlx4_en_cq *cq = priv->rx_cq[i];
2007 
2008 		napi_synchronize(&cq->napi);
2009 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
2010 		mlx4_en_deactivate_cq(priv, cq);
2011 
2012 		mlx4_en_free_affinity_hint(priv, i);
2013 	}
2014 }
2015 
2016 static void mlx4_en_restart(struct work_struct *work)
2017 {
2018 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2019 						 restart_task);
2020 	struct mlx4_en_dev *mdev = priv->mdev;
2021 	struct net_device *dev = priv->dev;
2022 
2023 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
2024 
2025 	rtnl_lock();
2026 	mutex_lock(&mdev->state_lock);
2027 	if (priv->port_up) {
2028 		mlx4_en_stop_port(dev, 1);
2029 		if (mlx4_en_start_port(dev))
2030 			en_err(priv, "Failed restarting port %d\n", priv->port);
2031 	}
2032 	mutex_unlock(&mdev->state_lock);
2033 	rtnl_unlock();
2034 }
2035 
2036 static void mlx4_en_clear_stats(struct net_device *dev)
2037 {
2038 	struct mlx4_en_priv *priv = netdev_priv(dev);
2039 	struct mlx4_en_dev *mdev = priv->mdev;
2040 	struct mlx4_en_tx_ring **tx_ring;
2041 	int i;
2042 
2043 	if (!mlx4_is_slave(mdev->dev))
2044 		if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
2045 			en_dbg(HW, priv, "Failed dumping statistics\n");
2046 
2047 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
2048 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
2049 	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
2050 	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
2051 	memset(&priv->rx_priority_flowstats, 0,
2052 	       sizeof(priv->rx_priority_flowstats));
2053 	memset(&priv->tx_priority_flowstats, 0,
2054 	       sizeof(priv->tx_priority_flowstats));
2055 	memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
2056 
2057 	tx_ring = priv->tx_ring[TX];
2058 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
2059 		tx_ring[i]->bytes = 0;
2060 		tx_ring[i]->packets = 0;
2061 		tx_ring[i]->tx_csum = 0;
2062 		tx_ring[i]->tx_dropped = 0;
2063 		tx_ring[i]->queue_stopped = 0;
2064 		tx_ring[i]->wake_queue = 0;
2065 		tx_ring[i]->tso_packets = 0;
2066 		tx_ring[i]->xmit_more = 0;
2067 	}
2068 	for (i = 0; i < priv->rx_ring_num; i++) {
2069 		priv->rx_ring[i]->bytes = 0;
2070 		priv->rx_ring[i]->packets = 0;
2071 		priv->rx_ring[i]->csum_ok = 0;
2072 		priv->rx_ring[i]->csum_none = 0;
2073 		priv->rx_ring[i]->csum_complete = 0;
2074 	}
2075 }
2076 
2077 static int mlx4_en_open(struct net_device *dev)
2078 {
2079 	struct mlx4_en_priv *priv = netdev_priv(dev);
2080 	struct mlx4_en_dev *mdev = priv->mdev;
2081 	int err = 0;
2082 
2083 	mutex_lock(&mdev->state_lock);
2084 
2085 	if (!mdev->device_up) {
2086 		en_err(priv, "Cannot open - device down/disabled\n");
2087 		err = -EBUSY;
2088 		goto out;
2089 	}
2090 
2091 	/* Reset HW statistics and SW counters */
2092 	mlx4_en_clear_stats(dev);
2093 
2094 	err = mlx4_en_start_port(dev);
2095 	if (err) {
2096 		en_err(priv, "Failed starting port:%d\n", priv->port);
2097 		goto out;
2098 	}
2099 	mlx4_en_linkstate(priv);
2100 out:
2101 	mutex_unlock(&mdev->state_lock);
2102 	return err;
2103 }
2104 
2105 
2106 static int mlx4_en_close(struct net_device *dev)
2107 {
2108 	struct mlx4_en_priv *priv = netdev_priv(dev);
2109 	struct mlx4_en_dev *mdev = priv->mdev;
2110 
2111 	en_dbg(IFDOWN, priv, "Close port called\n");
2112 
2113 	mutex_lock(&mdev->state_lock);
2114 
2115 	mlx4_en_stop_port(dev, 0);
2116 	netif_carrier_off(dev);
2117 
2118 	mutex_unlock(&mdev->state_lock);
2119 	return 0;
2120 }
2121 
2122 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2123 {
2124 	int i, t;
2125 
2126 #ifdef CONFIG_RFS_ACCEL
2127 	priv->dev->rx_cpu_rmap = NULL;
2128 #endif
2129 
2130 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2131 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2132 			if (priv->tx_ring[t] && priv->tx_ring[t][i])
2133 				mlx4_en_destroy_tx_ring(priv,
2134 							&priv->tx_ring[t][i]);
2135 			if (priv->tx_cq[t] && priv->tx_cq[t][i])
2136 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2137 		}
2138 		kfree(priv->tx_ring[t]);
2139 		kfree(priv->tx_cq[t]);
2140 	}
2141 
2142 	for (i = 0; i < priv->rx_ring_num; i++) {
2143 		if (priv->rx_ring[i])
2144 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2145 				priv->prof->rx_ring_size, priv->stride);
2146 		if (priv->rx_cq[i])
2147 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2148 	}
2149 
2150 }
2151 
2152 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2153 {
2154 	struct mlx4_en_port_profile *prof = priv->prof;
2155 	int i, t;
2156 	int node;
2157 
2158 	/* Create tx Rings */
2159 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2160 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2161 			node = cpu_to_node(i % num_online_cpus());
2162 			if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2163 					      prof->tx_ring_size, i, t, node))
2164 				goto err;
2165 
2166 			if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2167 						   prof->tx_ring_size,
2168 						   TXBB_SIZE, node, i))
2169 				goto err;
2170 		}
2171 	}
2172 
2173 	/* Create rx Rings */
2174 	for (i = 0; i < priv->rx_ring_num; i++) {
2175 		node = cpu_to_node(i % num_online_cpus());
2176 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2177 				      prof->rx_ring_size, i, RX, node))
2178 			goto err;
2179 
2180 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2181 					   prof->rx_ring_size, priv->stride,
2182 					   node, i))
2183 			goto err;
2184 
2185 	}
2186 
2187 #ifdef CONFIG_RFS_ACCEL
2188 	priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2189 #endif
2190 
2191 	return 0;
2192 
2193 err:
2194 	en_err(priv, "Failed to allocate NIC resources\n");
2195 	for (i = 0; i < priv->rx_ring_num; i++) {
2196 		if (priv->rx_ring[i])
2197 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2198 						prof->rx_ring_size,
2199 						priv->stride);
2200 		if (priv->rx_cq[i])
2201 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2202 	}
2203 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2204 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2205 			if (priv->tx_ring[t][i])
2206 				mlx4_en_destroy_tx_ring(priv,
2207 							&priv->tx_ring[t][i]);
2208 			if (priv->tx_cq[t][i])
2209 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2210 		}
2211 	}
2212 	return -ENOMEM;
2213 }
2214 
2215 
2216 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2217 			     struct mlx4_en_priv *src,
2218 			     struct mlx4_en_port_profile *prof)
2219 {
2220 	int t;
2221 
2222 	memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2223 	       sizeof(dst->hwtstamp_config));
2224 	dst->num_tx_rings_p_up = prof->num_tx_rings_p_up;
2225 	dst->rx_ring_num = prof->rx_ring_num;
2226 	dst->flags = prof->flags;
2227 	dst->mdev = src->mdev;
2228 	dst->port = src->port;
2229 	dst->dev = src->dev;
2230 	dst->prof = prof;
2231 	dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2232 					 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2233 
2234 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2235 		dst->tx_ring_num[t] = prof->tx_ring_num[t];
2236 		if (!dst->tx_ring_num[t])
2237 			continue;
2238 
2239 		dst->tx_ring[t] = kcalloc(MAX_TX_RINGS,
2240 					  sizeof(struct mlx4_en_tx_ring *),
2241 					  GFP_KERNEL);
2242 		if (!dst->tx_ring[t])
2243 			goto err_free_tx;
2244 
2245 		dst->tx_cq[t] = kcalloc(MAX_TX_RINGS,
2246 					sizeof(struct mlx4_en_cq *),
2247 					GFP_KERNEL);
2248 		if (!dst->tx_cq[t]) {
2249 			kfree(dst->tx_ring[t]);
2250 			goto err_free_tx;
2251 		}
2252 	}
2253 
2254 	return 0;
2255 
2256 err_free_tx:
2257 	while (t--) {
2258 		kfree(dst->tx_ring[t]);
2259 		kfree(dst->tx_cq[t]);
2260 	}
2261 	return -ENOMEM;
2262 }
2263 
2264 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2265 				struct mlx4_en_priv *src)
2266 {
2267 	int t;
2268 	memcpy(dst->rx_ring, src->rx_ring,
2269 	       sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2270 	memcpy(dst->rx_cq, src->rx_cq,
2271 	       sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2272 	memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2273 	       sizeof(dst->hwtstamp_config));
2274 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2275 		dst->tx_ring_num[t] = src->tx_ring_num[t];
2276 		dst->tx_ring[t] = src->tx_ring[t];
2277 		dst->tx_cq[t] = src->tx_cq[t];
2278 	}
2279 	dst->num_tx_rings_p_up = src->num_tx_rings_p_up;
2280 	dst->rx_ring_num = src->rx_ring_num;
2281 	memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2282 }
2283 
2284 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2285 				struct mlx4_en_priv *tmp,
2286 				struct mlx4_en_port_profile *prof,
2287 				bool carry_xdp_prog)
2288 {
2289 	struct bpf_prog *xdp_prog;
2290 	int i, t, ret;
2291 
2292 	ret = mlx4_en_copy_priv(tmp, priv, prof);
2293 	if (ret) {
2294 		en_warn(priv, "%s: mlx4_en_copy_priv() failed, return\n",
2295 			__func__);
2296 		return ret;
2297 	}
2298 
2299 	if (mlx4_en_alloc_resources(tmp)) {
2300 		en_warn(priv,
2301 			"%s: Resource allocation failed, using previous configuration\n",
2302 			__func__);
2303 		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2304 			kfree(tmp->tx_ring[t]);
2305 			kfree(tmp->tx_cq[t]);
2306 		}
2307 		return -ENOMEM;
2308 	}
2309 
2310 	/* All rx_rings has the same xdp_prog.  Pick the first one. */
2311 	xdp_prog = rcu_dereference_protected(
2312 		priv->rx_ring[0]->xdp_prog,
2313 		lockdep_is_held(&priv->mdev->state_lock));
2314 
2315 	if (xdp_prog && carry_xdp_prog) {
2316 		bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2317 		for (i = 0; i < tmp->rx_ring_num; i++)
2318 			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2319 					   xdp_prog);
2320 	}
2321 
2322 	return 0;
2323 }
2324 
2325 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2326 				    struct mlx4_en_priv *tmp)
2327 {
2328 	mlx4_en_free_resources(priv);
2329 	mlx4_en_update_priv(priv, tmp);
2330 }
2331 
2332 void mlx4_en_destroy_netdev(struct net_device *dev)
2333 {
2334 	struct mlx4_en_priv *priv = netdev_priv(dev);
2335 	struct mlx4_en_dev *mdev = priv->mdev;
2336 
2337 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2338 
2339 	/* Unregister device - this will close the port if it was up */
2340 	if (priv->registered) {
2341 		devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2342 							      priv->port));
2343 		unregister_netdev(dev);
2344 	}
2345 
2346 	if (priv->allocated)
2347 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2348 
2349 	cancel_delayed_work(&priv->stats_task);
2350 	cancel_delayed_work(&priv->service_task);
2351 	/* flush any pending task for this netdev */
2352 	flush_workqueue(mdev->workqueue);
2353 
2354 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2355 		mlx4_en_remove_timestamp(mdev);
2356 
2357 	/* Detach the netdev so tasks would not attempt to access it */
2358 	mutex_lock(&mdev->state_lock);
2359 	mdev->pndev[priv->port] = NULL;
2360 	mdev->upper[priv->port] = NULL;
2361 
2362 #ifdef CONFIG_RFS_ACCEL
2363 	mlx4_en_cleanup_filters(priv);
2364 #endif
2365 
2366 	mlx4_en_free_resources(priv);
2367 	mutex_unlock(&mdev->state_lock);
2368 
2369 	free_netdev(dev);
2370 }
2371 
2372 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2373 {
2374 	struct mlx4_en_priv *priv = netdev_priv(dev);
2375 
2376 	if (mtu > MLX4_EN_MAX_XDP_MTU) {
2377 		en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2378 		       mtu, MLX4_EN_MAX_XDP_MTU);
2379 		return false;
2380 	}
2381 
2382 	return true;
2383 }
2384 
2385 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2386 {
2387 	struct mlx4_en_priv *priv = netdev_priv(dev);
2388 	struct mlx4_en_dev *mdev = priv->mdev;
2389 	int err = 0;
2390 
2391 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2392 		 dev->mtu, new_mtu);
2393 
2394 	if (priv->tx_ring_num[TX_XDP] &&
2395 	    !mlx4_en_check_xdp_mtu(dev, new_mtu))
2396 		return -EOPNOTSUPP;
2397 
2398 	dev->mtu = new_mtu;
2399 
2400 	if (netif_running(dev)) {
2401 		mutex_lock(&mdev->state_lock);
2402 		if (!mdev->device_up) {
2403 			/* NIC is probably restarting - let restart task reset
2404 			 * the port */
2405 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2406 		} else {
2407 			mlx4_en_stop_port(dev, 1);
2408 			err = mlx4_en_start_port(dev);
2409 			if (err) {
2410 				en_err(priv, "Failed restarting port:%d\n",
2411 					 priv->port);
2412 				if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING,
2413 						      &priv->state))
2414 					queue_work(mdev->workqueue, &priv->restart_task);
2415 			}
2416 		}
2417 		mutex_unlock(&mdev->state_lock);
2418 	}
2419 	return 0;
2420 }
2421 
2422 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2423 {
2424 	struct mlx4_en_priv *priv = netdev_priv(dev);
2425 	struct mlx4_en_dev *mdev = priv->mdev;
2426 	struct hwtstamp_config config;
2427 
2428 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2429 		return -EFAULT;
2430 
2431 	/* device doesn't support time stamping */
2432 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2433 		return -EINVAL;
2434 
2435 	/* TX HW timestamp */
2436 	switch (config.tx_type) {
2437 	case HWTSTAMP_TX_OFF:
2438 	case HWTSTAMP_TX_ON:
2439 		break;
2440 	default:
2441 		return -ERANGE;
2442 	}
2443 
2444 	/* RX HW timestamp */
2445 	switch (config.rx_filter) {
2446 	case HWTSTAMP_FILTER_NONE:
2447 		break;
2448 	case HWTSTAMP_FILTER_ALL:
2449 	case HWTSTAMP_FILTER_SOME:
2450 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2451 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2452 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2453 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2454 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2455 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2456 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2457 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2458 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2459 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2460 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2461 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2462 	case HWTSTAMP_FILTER_NTP_ALL:
2463 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2464 		break;
2465 	default:
2466 		return -ERANGE;
2467 	}
2468 
2469 	if (mlx4_en_reset_config(dev, config, dev->features)) {
2470 		config.tx_type = HWTSTAMP_TX_OFF;
2471 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2472 	}
2473 
2474 	return copy_to_user(ifr->ifr_data, &config,
2475 			    sizeof(config)) ? -EFAULT : 0;
2476 }
2477 
2478 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2479 {
2480 	struct mlx4_en_priv *priv = netdev_priv(dev);
2481 
2482 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2483 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2484 }
2485 
2486 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2487 {
2488 	switch (cmd) {
2489 	case SIOCSHWTSTAMP:
2490 		return mlx4_en_hwtstamp_set(dev, ifr);
2491 	case SIOCGHWTSTAMP:
2492 		return mlx4_en_hwtstamp_get(dev, ifr);
2493 	default:
2494 		return -EOPNOTSUPP;
2495 	}
2496 }
2497 
2498 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2499 					      netdev_features_t features)
2500 {
2501 	struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2502 	struct mlx4_en_dev *mdev = en_priv->mdev;
2503 
2504 	/* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2505 	 * enable/disable make sure S-TAG flag is always in same state as
2506 	 * C-TAG.
2507 	 */
2508 	if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2509 	    !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2510 		features |= NETIF_F_HW_VLAN_STAG_RX;
2511 	else
2512 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2513 
2514 	return features;
2515 }
2516 
2517 static int mlx4_en_set_features(struct net_device *netdev,
2518 		netdev_features_t features)
2519 {
2520 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2521 	bool reset = false;
2522 	int ret = 0;
2523 
2524 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2525 		en_info(priv, "Turn %s RX-FCS\n",
2526 			(features & NETIF_F_RXFCS) ? "ON" : "OFF");
2527 		reset = true;
2528 	}
2529 
2530 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2531 		u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2532 
2533 		en_info(priv, "Turn %s RX-ALL\n",
2534 			ignore_fcs_value ? "ON" : "OFF");
2535 		ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2536 					      priv->port, ignore_fcs_value);
2537 		if (ret)
2538 			return ret;
2539 	}
2540 
2541 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2542 		en_info(priv, "Turn %s RX vlan strip offload\n",
2543 			(features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2544 		reset = true;
2545 	}
2546 
2547 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2548 		en_info(priv, "Turn %s TX vlan strip offload\n",
2549 			(features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2550 
2551 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2552 		en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2553 			(features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2554 
2555 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2556 		en_info(priv, "Turn %s loopback\n",
2557 			(features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2558 		mlx4_en_update_loopback_state(netdev, features);
2559 	}
2560 
2561 	if (reset) {
2562 		ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2563 					   features);
2564 		if (ret)
2565 			return ret;
2566 	}
2567 
2568 	return 0;
2569 }
2570 
2571 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2572 {
2573 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2574 	struct mlx4_en_dev *mdev = en_priv->mdev;
2575 
2576 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac);
2577 }
2578 
2579 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2580 			       __be16 vlan_proto)
2581 {
2582 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2583 	struct mlx4_en_dev *mdev = en_priv->mdev;
2584 
2585 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2586 				vlan_proto);
2587 }
2588 
2589 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2590 			       int max_tx_rate)
2591 {
2592 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2593 	struct mlx4_en_dev *mdev = en_priv->mdev;
2594 
2595 	return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2596 				max_tx_rate);
2597 }
2598 
2599 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2600 {
2601 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2602 	struct mlx4_en_dev *mdev = en_priv->mdev;
2603 
2604 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2605 }
2606 
2607 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2608 {
2609 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2610 	struct mlx4_en_dev *mdev = en_priv->mdev;
2611 
2612 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2613 }
2614 
2615 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2616 {
2617 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2618 	struct mlx4_en_dev *mdev = en_priv->mdev;
2619 
2620 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2621 }
2622 
2623 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2624 				struct ifla_vf_stats *vf_stats)
2625 {
2626 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2627 	struct mlx4_en_dev *mdev = en_priv->mdev;
2628 
2629 	return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2630 }
2631 
2632 #define PORT_ID_BYTE_LEN 8
2633 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2634 				    struct netdev_phys_item_id *ppid)
2635 {
2636 	struct mlx4_en_priv *priv = netdev_priv(dev);
2637 	struct mlx4_dev *mdev = priv->mdev->dev;
2638 	int i;
2639 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2640 
2641 	if (!phys_port_id)
2642 		return -EOPNOTSUPP;
2643 
2644 	ppid->id_len = sizeof(phys_port_id);
2645 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2646 		ppid->id[i] =  phys_port_id & 0xff;
2647 		phys_port_id >>= 8;
2648 	}
2649 	return 0;
2650 }
2651 
2652 static int mlx4_udp_tunnel_sync(struct net_device *dev, unsigned int table)
2653 {
2654 	struct mlx4_en_priv *priv = netdev_priv(dev);
2655 	struct udp_tunnel_info ti;
2656 	int ret;
2657 
2658 	udp_tunnel_nic_get_port(dev, table, 0, &ti);
2659 	priv->vxlan_port = ti.port;
2660 
2661 	ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2662 	if (ret)
2663 		return ret;
2664 
2665 	return mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2666 				   VXLAN_STEER_BY_OUTER_MAC,
2667 				   !!priv->vxlan_port);
2668 }
2669 
2670 static const struct udp_tunnel_nic_info mlx4_udp_tunnels = {
2671 	.sync_table	= mlx4_udp_tunnel_sync,
2672 	.flags		= UDP_TUNNEL_NIC_INFO_MAY_SLEEP |
2673 			  UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
2674 	.tables		= {
2675 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
2676 	},
2677 };
2678 
2679 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2680 						struct net_device *dev,
2681 						netdev_features_t features)
2682 {
2683 	features = vlan_features_check(skb, features);
2684 	features = vxlan_features_check(skb, features);
2685 
2686 	/* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2687 	 * support inner IPv6 checksums and segmentation so  we need to
2688 	 * strip that feature if this is an IPv6 encapsulated frame.
2689 	 */
2690 	if (skb->encapsulation &&
2691 	    (skb->ip_summed == CHECKSUM_PARTIAL)) {
2692 		struct mlx4_en_priv *priv = netdev_priv(dev);
2693 
2694 		if (!priv->vxlan_port ||
2695 		    (ip_hdr(skb)->version != 4) ||
2696 		    (udp_hdr(skb)->dest != priv->vxlan_port))
2697 			features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2698 	}
2699 
2700 	return features;
2701 }
2702 
2703 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2704 {
2705 	struct mlx4_en_priv *priv = netdev_priv(dev);
2706 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2707 	struct mlx4_update_qp_params params;
2708 	int err;
2709 
2710 	if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2711 		return -EOPNOTSUPP;
2712 
2713 	/* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2714 	if (maxrate >> 12) {
2715 		params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2716 		params.rate_val  = maxrate / 1000;
2717 	} else if (maxrate) {
2718 		params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2719 		params.rate_val  = maxrate;
2720 	} else { /* zero serves to revoke the QP rate-limitation */
2721 		params.rate_unit = 0;
2722 		params.rate_val  = 0;
2723 	}
2724 
2725 	err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2726 			     &params);
2727 	return err;
2728 }
2729 
2730 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2731 {
2732 	struct mlx4_en_priv *priv = netdev_priv(dev);
2733 	struct mlx4_en_dev *mdev = priv->mdev;
2734 	struct mlx4_en_port_profile new_prof;
2735 	struct bpf_prog *old_prog;
2736 	struct mlx4_en_priv *tmp;
2737 	int tx_changed = 0;
2738 	int xdp_ring_num;
2739 	int port_up = 0;
2740 	int err;
2741 	int i;
2742 
2743 	xdp_ring_num = prog ? priv->rx_ring_num : 0;
2744 
2745 	/* No need to reconfigure buffers when simply swapping the
2746 	 * program for a new one.
2747 	 */
2748 	if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2749 		if (prog)
2750 			bpf_prog_add(prog, priv->rx_ring_num - 1);
2751 
2752 		mutex_lock(&mdev->state_lock);
2753 		for (i = 0; i < priv->rx_ring_num; i++) {
2754 			old_prog = rcu_dereference_protected(
2755 					priv->rx_ring[i]->xdp_prog,
2756 					lockdep_is_held(&mdev->state_lock));
2757 			rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2758 			if (old_prog)
2759 				bpf_prog_put(old_prog);
2760 		}
2761 		mutex_unlock(&mdev->state_lock);
2762 		return 0;
2763 	}
2764 
2765 	if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2766 		return -EOPNOTSUPP;
2767 
2768 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2769 	if (!tmp)
2770 		return -ENOMEM;
2771 
2772 	if (prog)
2773 		bpf_prog_add(prog, priv->rx_ring_num - 1);
2774 
2775 	mutex_lock(&mdev->state_lock);
2776 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2777 	new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2778 
2779 	if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2780 		tx_changed = 1;
2781 		new_prof.tx_ring_num[TX] =
2782 			MAX_TX_RINGS - ALIGN(xdp_ring_num, priv->prof->num_up);
2783 		en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2784 	}
2785 
2786 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2787 	if (err) {
2788 		if (prog)
2789 			bpf_prog_sub(prog, priv->rx_ring_num - 1);
2790 		goto unlock_out;
2791 	}
2792 
2793 	if (priv->port_up) {
2794 		port_up = 1;
2795 		mlx4_en_stop_port(dev, 1);
2796 	}
2797 
2798 	mlx4_en_safe_replace_resources(priv, tmp);
2799 	if (tx_changed)
2800 		netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2801 
2802 	for (i = 0; i < priv->rx_ring_num; i++) {
2803 		old_prog = rcu_dereference_protected(
2804 					priv->rx_ring[i]->xdp_prog,
2805 					lockdep_is_held(&mdev->state_lock));
2806 		rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2807 		if (old_prog)
2808 			bpf_prog_put(old_prog);
2809 	}
2810 
2811 	if (port_up) {
2812 		err = mlx4_en_start_port(dev);
2813 		if (err) {
2814 			en_err(priv, "Failed starting port %d for XDP change\n",
2815 			       priv->port);
2816 			if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state))
2817 				queue_work(mdev->workqueue, &priv->restart_task);
2818 		}
2819 	}
2820 
2821 unlock_out:
2822 	mutex_unlock(&mdev->state_lock);
2823 	kfree(tmp);
2824 	return err;
2825 }
2826 
2827 static int mlx4_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2828 {
2829 	switch (xdp->command) {
2830 	case XDP_SETUP_PROG:
2831 		return mlx4_xdp_set(dev, xdp->prog);
2832 	default:
2833 		return -EINVAL;
2834 	}
2835 }
2836 
2837 static const struct net_device_ops mlx4_netdev_ops = {
2838 	.ndo_open		= mlx4_en_open,
2839 	.ndo_stop		= mlx4_en_close,
2840 	.ndo_start_xmit		= mlx4_en_xmit,
2841 	.ndo_select_queue	= mlx4_en_select_queue,
2842 	.ndo_get_stats64	= mlx4_en_get_stats64,
2843 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2844 	.ndo_set_mac_address	= mlx4_en_set_mac,
2845 	.ndo_validate_addr	= eth_validate_addr,
2846 	.ndo_change_mtu		= mlx4_en_change_mtu,
2847 	.ndo_eth_ioctl		= mlx4_en_ioctl,
2848 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2849 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2850 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2851 	.ndo_set_features	= mlx4_en_set_features,
2852 	.ndo_fix_features	= mlx4_en_fix_features,
2853 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2854 #ifdef CONFIG_RFS_ACCEL
2855 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2856 #endif
2857 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2858 	.ndo_features_check	= mlx4_en_features_check,
2859 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2860 	.ndo_bpf		= mlx4_xdp,
2861 };
2862 
2863 static const struct net_device_ops mlx4_netdev_ops_master = {
2864 	.ndo_open		= mlx4_en_open,
2865 	.ndo_stop		= mlx4_en_close,
2866 	.ndo_start_xmit		= mlx4_en_xmit,
2867 	.ndo_select_queue	= mlx4_en_select_queue,
2868 	.ndo_get_stats64	= mlx4_en_get_stats64,
2869 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2870 	.ndo_set_mac_address	= mlx4_en_set_mac,
2871 	.ndo_validate_addr	= eth_validate_addr,
2872 	.ndo_change_mtu		= mlx4_en_change_mtu,
2873 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2874 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2875 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2876 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2877 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2878 	.ndo_set_vf_rate	= mlx4_en_set_vf_rate,
2879 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2880 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2881 	.ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2882 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2883 	.ndo_set_features	= mlx4_en_set_features,
2884 	.ndo_fix_features	= mlx4_en_fix_features,
2885 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2886 #ifdef CONFIG_RFS_ACCEL
2887 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2888 #endif
2889 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2890 	.ndo_features_check	= mlx4_en_features_check,
2891 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2892 	.ndo_bpf		= mlx4_xdp,
2893 };
2894 
2895 struct mlx4_en_bond {
2896 	struct work_struct work;
2897 	struct mlx4_en_priv *priv;
2898 	int is_bonded;
2899 	struct mlx4_port_map port_map;
2900 };
2901 
2902 static void mlx4_en_bond_work(struct work_struct *work)
2903 {
2904 	struct mlx4_en_bond *bond = container_of(work,
2905 						     struct mlx4_en_bond,
2906 						     work);
2907 	int err = 0;
2908 	struct mlx4_dev *dev = bond->priv->mdev->dev;
2909 
2910 	if (bond->is_bonded) {
2911 		if (!mlx4_is_bonded(dev)) {
2912 			err = mlx4_bond(dev);
2913 			if (err)
2914 				en_err(bond->priv, "Fail to bond device\n");
2915 		}
2916 		if (!err) {
2917 			err = mlx4_port_map_set(dev, &bond->port_map);
2918 			if (err)
2919 				en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2920 				       bond->port_map.port1,
2921 				       bond->port_map.port2,
2922 				       err);
2923 		}
2924 	} else if (mlx4_is_bonded(dev)) {
2925 		err = mlx4_unbond(dev);
2926 		if (err)
2927 			en_err(bond->priv, "Fail to unbond device\n");
2928 	}
2929 	dev_put(bond->priv->dev);
2930 	kfree(bond);
2931 }
2932 
2933 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2934 				   u8 v2p_p1, u8 v2p_p2)
2935 {
2936 	struct mlx4_en_bond *bond = NULL;
2937 
2938 	bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2939 	if (!bond)
2940 		return -ENOMEM;
2941 
2942 	INIT_WORK(&bond->work, mlx4_en_bond_work);
2943 	bond->priv = priv;
2944 	bond->is_bonded = is_bonded;
2945 	bond->port_map.port1 = v2p_p1;
2946 	bond->port_map.port2 = v2p_p2;
2947 	dev_hold(priv->dev);
2948 	queue_work(priv->mdev->workqueue, &bond->work);
2949 	return 0;
2950 }
2951 
2952 int mlx4_en_netdev_event(struct notifier_block *this,
2953 			 unsigned long event, void *ptr)
2954 {
2955 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2956 	u8 port = 0;
2957 	struct mlx4_en_dev *mdev;
2958 	struct mlx4_dev *dev;
2959 	int i, num_eth_ports = 0;
2960 	bool do_bond = true;
2961 	struct mlx4_en_priv *priv;
2962 	u8 v2p_port1 = 0;
2963 	u8 v2p_port2 = 0;
2964 
2965 	if (!net_eq(dev_net(ndev), &init_net))
2966 		return NOTIFY_DONE;
2967 
2968 	mdev = container_of(this, struct mlx4_en_dev, nb);
2969 	dev = mdev->dev;
2970 
2971 	/* Go into this mode only when two network devices set on two ports
2972 	 * of the same mlx4 device are slaves of the same bonding master
2973 	 */
2974 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2975 		++num_eth_ports;
2976 		if (!port && (mdev->pndev[i] == ndev))
2977 			port = i;
2978 		mdev->upper[i] = mdev->pndev[i] ?
2979 			netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2980 		/* condition not met: network device is a slave */
2981 		if (!mdev->upper[i])
2982 			do_bond = false;
2983 		if (num_eth_ports < 2)
2984 			continue;
2985 		/* condition not met: same master */
2986 		if (mdev->upper[i] != mdev->upper[i-1])
2987 			do_bond = false;
2988 	}
2989 	/* condition not met: 2 salves */
2990 	do_bond = (num_eth_ports ==  2) ? do_bond : false;
2991 
2992 	/* handle only events that come with enough info */
2993 	if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
2994 		return NOTIFY_DONE;
2995 
2996 	priv = netdev_priv(ndev);
2997 	if (do_bond) {
2998 		struct netdev_notifier_bonding_info *notifier_info = ptr;
2999 		struct netdev_bonding_info *bonding_info =
3000 			&notifier_info->bonding_info;
3001 
3002 		/* required mode 1, 2 or 4 */
3003 		if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3004 		    (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
3005 		    (bonding_info->master.bond_mode != BOND_MODE_8023AD))
3006 			do_bond = false;
3007 
3008 		/* require exactly 2 slaves */
3009 		if (bonding_info->master.num_slaves != 2)
3010 			do_bond = false;
3011 
3012 		/* calc v2p */
3013 		if (do_bond) {
3014 			if (bonding_info->master.bond_mode ==
3015 			    BOND_MODE_ACTIVEBACKUP) {
3016 				/* in active-backup mode virtual ports are
3017 				 * mapped to the physical port of the active
3018 				 * slave */
3019 				if (bonding_info->slave.state ==
3020 				    BOND_STATE_BACKUP) {
3021 					if (port == 1) {
3022 						v2p_port1 = 2;
3023 						v2p_port2 = 2;
3024 					} else {
3025 						v2p_port1 = 1;
3026 						v2p_port2 = 1;
3027 					}
3028 				} else { /* BOND_STATE_ACTIVE */
3029 					if (port == 1) {
3030 						v2p_port1 = 1;
3031 						v2p_port2 = 1;
3032 					} else {
3033 						v2p_port1 = 2;
3034 						v2p_port2 = 2;
3035 					}
3036 				}
3037 			} else { /* Active-Active */
3038 				/* in active-active mode a virtual port is
3039 				 * mapped to the native physical port if and only
3040 				 * if the physical port is up */
3041 				__s8 link = bonding_info->slave.link;
3042 
3043 				if (port == 1)
3044 					v2p_port2 = 2;
3045 				else
3046 					v2p_port1 = 1;
3047 				if ((link == BOND_LINK_UP) ||
3048 				    (link == BOND_LINK_FAIL)) {
3049 					if (port == 1)
3050 						v2p_port1 = 1;
3051 					else
3052 						v2p_port2 = 2;
3053 				} else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3054 					if (port == 1)
3055 						v2p_port1 = 2;
3056 					else
3057 						v2p_port2 = 1;
3058 				}
3059 			}
3060 		}
3061 	}
3062 
3063 	mlx4_en_queue_bond_work(priv, do_bond,
3064 				v2p_port1, v2p_port2);
3065 
3066 	return NOTIFY_DONE;
3067 }
3068 
3069 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3070 				     struct mlx4_en_stats_bitmap *stats_bitmap,
3071 				     u8 rx_ppp, u8 rx_pause,
3072 				     u8 tx_ppp, u8 tx_pause)
3073 {
3074 	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3075 
3076 	if (!mlx4_is_slave(dev) &&
3077 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3078 		mutex_lock(&stats_bitmap->mutex);
3079 		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3080 
3081 		if (rx_ppp)
3082 			bitmap_set(stats_bitmap->bitmap, last_i,
3083 				   NUM_FLOW_PRIORITY_STATS_RX);
3084 		last_i += NUM_FLOW_PRIORITY_STATS_RX;
3085 
3086 		if (rx_pause && !(rx_ppp))
3087 			bitmap_set(stats_bitmap->bitmap, last_i,
3088 				   NUM_FLOW_STATS_RX);
3089 		last_i += NUM_FLOW_STATS_RX;
3090 
3091 		if (tx_ppp)
3092 			bitmap_set(stats_bitmap->bitmap, last_i,
3093 				   NUM_FLOW_PRIORITY_STATS_TX);
3094 		last_i += NUM_FLOW_PRIORITY_STATS_TX;
3095 
3096 		if (tx_pause && !(tx_ppp))
3097 			bitmap_set(stats_bitmap->bitmap, last_i,
3098 				   NUM_FLOW_STATS_TX);
3099 		last_i += NUM_FLOW_STATS_TX;
3100 
3101 		mutex_unlock(&stats_bitmap->mutex);
3102 	}
3103 }
3104 
3105 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3106 			      struct mlx4_en_stats_bitmap *stats_bitmap,
3107 			      u8 rx_ppp, u8 rx_pause,
3108 			      u8 tx_ppp, u8 tx_pause)
3109 {
3110 	int last_i = 0;
3111 
3112 	mutex_init(&stats_bitmap->mutex);
3113 	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3114 
3115 	if (mlx4_is_slave(dev)) {
3116 		bitmap_set(stats_bitmap->bitmap, last_i +
3117 					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3118 		bitmap_set(stats_bitmap->bitmap, last_i +
3119 					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3120 		bitmap_set(stats_bitmap->bitmap, last_i +
3121 					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3122 		bitmap_set(stats_bitmap->bitmap, last_i +
3123 					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3124 		bitmap_set(stats_bitmap->bitmap, last_i +
3125 					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3126 		bitmap_set(stats_bitmap->bitmap, last_i +
3127 					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3128 	} else {
3129 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3130 	}
3131 	last_i += NUM_MAIN_STATS;
3132 
3133 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3134 	last_i += NUM_PORT_STATS;
3135 
3136 	if (mlx4_is_master(dev))
3137 		bitmap_set(stats_bitmap->bitmap, last_i,
3138 			   NUM_PF_STATS);
3139 	last_i += NUM_PF_STATS;
3140 
3141 	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3142 					rx_ppp, rx_pause,
3143 					tx_ppp, tx_pause);
3144 	last_i += NUM_FLOW_STATS;
3145 
3146 	if (!mlx4_is_slave(dev))
3147 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3148 	last_i += NUM_PKT_STATS;
3149 
3150 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3151 	last_i += NUM_XDP_STATS;
3152 
3153 	if (!mlx4_is_slave(dev))
3154 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PHY_STATS);
3155 	last_i += NUM_PHY_STATS;
3156 }
3157 
3158 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3159 			struct mlx4_en_port_profile *prof)
3160 {
3161 	struct net_device *dev;
3162 	struct mlx4_en_priv *priv;
3163 	int i, t;
3164 	int err;
3165 
3166 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3167 				 MAX_TX_RINGS, MAX_RX_RINGS);
3168 	if (dev == NULL)
3169 		return -ENOMEM;
3170 
3171 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3172 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3173 
3174 	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3175 	dev->dev_port = port - 1;
3176 
3177 	/*
3178 	 * Initialize driver private data
3179 	 */
3180 
3181 	priv = netdev_priv(dev);
3182 	memset(priv, 0, sizeof(struct mlx4_en_priv));
3183 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3184 	spin_lock_init(&priv->stats_lock);
3185 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3186 	INIT_WORK(&priv->restart_task, mlx4_en_restart);
3187 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate_work);
3188 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3189 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3190 #ifdef CONFIG_RFS_ACCEL
3191 	INIT_LIST_HEAD(&priv->filters);
3192 	spin_lock_init(&priv->filters_lock);
3193 #endif
3194 
3195 	priv->dev = dev;
3196 	priv->mdev = mdev;
3197 	priv->ddev = &mdev->pdev->dev;
3198 	priv->prof = prof;
3199 	priv->port = port;
3200 	priv->port_up = false;
3201 	priv->flags = prof->flags;
3202 	priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3203 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3204 			MLX4_WQE_CTRL_SOLICITED);
3205 	priv->num_tx_rings_p_up = mdev->profile.max_num_tx_rings_p_up;
3206 	priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3207 	netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3208 
3209 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3210 		priv->tx_ring_num[t] = prof->tx_ring_num[t];
3211 		if (!priv->tx_ring_num[t])
3212 			continue;
3213 
3214 		priv->tx_ring[t] = kcalloc(MAX_TX_RINGS,
3215 					   sizeof(struct mlx4_en_tx_ring *),
3216 					   GFP_KERNEL);
3217 		if (!priv->tx_ring[t]) {
3218 			err = -ENOMEM;
3219 			goto out;
3220 		}
3221 		priv->tx_cq[t] = kcalloc(MAX_TX_RINGS,
3222 					 sizeof(struct mlx4_en_cq *),
3223 					 GFP_KERNEL);
3224 		if (!priv->tx_cq[t]) {
3225 			err = -ENOMEM;
3226 			goto out;
3227 		}
3228 	}
3229 	priv->rx_ring_num = prof->rx_ring_num;
3230 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3231 	priv->cqe_size = mdev->dev->caps.cqe_size;
3232 	priv->mac_index = -1;
3233 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
3234 #ifdef CONFIG_MLX4_EN_DCB
3235 	if (!mlx4_is_slave(priv->mdev->dev)) {
3236 		u8 prio;
3237 
3238 		for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; ++prio) {
3239 			priv->ets.prio_tc[prio] = prio;
3240 			priv->ets.tc_tsa[prio]  = IEEE_8021QAZ_TSA_VENDOR;
3241 		}
3242 
3243 		priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3244 			DCB_CAP_DCBX_VER_IEEE;
3245 		priv->flags |= MLX4_EN_DCB_ENABLED;
3246 		priv->cee_config.pfc_state = false;
3247 
3248 		for (i = 0; i < MLX4_EN_NUM_UP_HIGH; i++)
3249 			priv->cee_config.dcb_pfc[i] = pfc_disabled;
3250 
3251 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3252 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3253 		} else {
3254 			en_info(priv, "enabling only PFC DCB ops\n");
3255 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3256 		}
3257 	}
3258 #endif
3259 
3260 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3261 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
3262 
3263 	/* Query for default mac and max mtu */
3264 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3265 
3266 	if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3267 	    MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3268 		priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3269 
3270 	/* Set default MAC */
3271 	dev->addr_len = ETH_ALEN;
3272 	mlx4_en_u64_to_mac(dev, mdev->dev->caps.def_mac[priv->port]);
3273 	if (!is_valid_ether_addr(dev->dev_addr)) {
3274 		en_err(priv, "Port: %d, invalid mac burned: %pM, quitting\n",
3275 		       priv->port, dev->dev_addr);
3276 		err = -EINVAL;
3277 		goto out;
3278 	} else if (mlx4_is_slave(priv->mdev->dev) &&
3279 		   (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3280 		/* Random MAC was assigned in mlx4_slave_cap
3281 		 * in mlx4_core module
3282 		 */
3283 		dev->addr_assign_type |= NET_ADDR_RANDOM;
3284 		en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3285 	}
3286 
3287 	memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3288 
3289 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3290 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3291 	err = mlx4_en_alloc_resources(priv);
3292 	if (err)
3293 		goto out;
3294 
3295 	/* Initialize time stamping config */
3296 	priv->hwtstamp_config.flags = 0;
3297 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3298 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3299 
3300 	/* Allocate page for receive rings */
3301 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3302 				MLX4_EN_PAGE_SIZE);
3303 	if (err) {
3304 		en_err(priv, "Failed to allocate page for rx qps\n");
3305 		goto out;
3306 	}
3307 	priv->allocated = 1;
3308 
3309 	/*
3310 	 * Initialize netdev entry points
3311 	 */
3312 	if (mlx4_is_master(priv->mdev->dev))
3313 		dev->netdev_ops = &mlx4_netdev_ops_master;
3314 	else
3315 		dev->netdev_ops = &mlx4_netdev_ops;
3316 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3317 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3318 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3319 
3320 	dev->ethtool_ops = &mlx4_en_ethtool_ops;
3321 
3322 	/*
3323 	 * Set driver features
3324 	 */
3325 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3326 	if (mdev->LSO_support)
3327 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3328 
3329 	if (mdev->dev->caps.tunnel_offload_mode ==
3330 	    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3331 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3332 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3333 				    NETIF_F_GSO_PARTIAL;
3334 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3335 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3336 				    NETIF_F_GSO_PARTIAL;
3337 		dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3338 		dev->hw_enc_features = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3339 				       NETIF_F_RXCSUM |
3340 				       NETIF_F_TSO | NETIF_F_TSO6 |
3341 				       NETIF_F_GSO_UDP_TUNNEL |
3342 				       NETIF_F_GSO_UDP_TUNNEL_CSUM |
3343 				       NETIF_F_GSO_PARTIAL;
3344 
3345 		dev->udp_tunnel_nic_info = &mlx4_udp_tunnels;
3346 	}
3347 
3348 	dev->vlan_features = dev->hw_features;
3349 
3350 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3351 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3352 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3353 			NETIF_F_HW_VLAN_CTAG_FILTER;
3354 	dev->hw_features |= NETIF_F_LOOPBACK |
3355 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3356 
3357 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3358 		dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3359 			NETIF_F_HW_VLAN_STAG_FILTER;
3360 		dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3361 	}
3362 
3363 	if (mlx4_is_slave(mdev->dev)) {
3364 		bool vlan_offload_disabled;
3365 		int phv;
3366 
3367 		err = get_phv_bit(mdev->dev, port, &phv);
3368 		if (!err && phv) {
3369 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3370 			priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3371 		}
3372 		err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3373 							&vlan_offload_disabled);
3374 		if (!err && vlan_offload_disabled) {
3375 			dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3376 					      NETIF_F_HW_VLAN_CTAG_RX |
3377 					      NETIF_F_HW_VLAN_STAG_TX |
3378 					      NETIF_F_HW_VLAN_STAG_RX);
3379 			dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3380 					   NETIF_F_HW_VLAN_CTAG_RX |
3381 					   NETIF_F_HW_VLAN_STAG_TX |
3382 					   NETIF_F_HW_VLAN_STAG_RX);
3383 		}
3384 	} else {
3385 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3386 		    !(mdev->dev->caps.flags2 &
3387 		      MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3388 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3389 	}
3390 
3391 	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3392 		dev->hw_features |= NETIF_F_RXFCS;
3393 
3394 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3395 		dev->hw_features |= NETIF_F_RXALL;
3396 
3397 	if (mdev->dev->caps.steering_mode ==
3398 	    MLX4_STEERING_MODE_DEVICE_MANAGED &&
3399 	    mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3400 		dev->hw_features |= NETIF_F_NTUPLE;
3401 
3402 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3403 		dev->priv_flags |= IFF_UNICAST_FLT;
3404 
3405 	/* Setting a default hash function value */
3406 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3407 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3408 	} else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3409 		priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3410 	} else {
3411 		en_warn(priv,
3412 			"No RSS hash capabilities exposed, using Toeplitz\n");
3413 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3414 	}
3415 
3416 	/* MTU range: 68 - hw-specific max */
3417 	dev->min_mtu = ETH_MIN_MTU;
3418 	dev->max_mtu = priv->max_mtu;
3419 
3420 	/* supports LSOv2 packets. */
3421 	netif_set_tso_max_size(dev, GSO_MAX_SIZE);
3422 
3423 	mdev->pndev[port] = dev;
3424 	mdev->upper[port] = NULL;
3425 
3426 	netif_carrier_off(dev);
3427 	mlx4_en_set_default_moderation(priv);
3428 
3429 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3430 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3431 
3432 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3433 
3434 	/* Configure port */
3435 	mlx4_en_calc_rx_buf(dev);
3436 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3437 				    priv->rx_skb_size + ETH_FCS_LEN,
3438 				    prof->tx_pause, prof->tx_ppp,
3439 				    prof->rx_pause, prof->rx_ppp);
3440 	if (err) {
3441 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3442 		       priv->port, err);
3443 		goto out;
3444 	}
3445 
3446 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3447 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3448 		if (err) {
3449 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3450 			       err);
3451 			goto out;
3452 		}
3453 	}
3454 
3455 	/* Init port */
3456 	en_warn(priv, "Initializing port\n");
3457 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
3458 	if (err) {
3459 		en_err(priv, "Failed Initializing port\n");
3460 		goto out;
3461 	}
3462 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3463 
3464 	/* Initialize time stamp mechanism */
3465 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3466 		mlx4_en_init_timestamp(mdev);
3467 
3468 	queue_delayed_work(mdev->workqueue, &priv->service_task,
3469 			   SERVICE_TASK_DELAY);
3470 
3471 	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3472 				 mdev->profile.prof[priv->port].rx_ppp,
3473 				 mdev->profile.prof[priv->port].rx_pause,
3474 				 mdev->profile.prof[priv->port].tx_ppp,
3475 				 mdev->profile.prof[priv->port].tx_pause);
3476 
3477 	err = register_netdev(dev);
3478 	if (err) {
3479 		en_err(priv, "Netdev registration failed for port %d\n", port);
3480 		goto out;
3481 	}
3482 
3483 	priv->registered = 1;
3484 	devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3485 				  dev);
3486 
3487 	return 0;
3488 
3489 out:
3490 	mlx4_en_destroy_netdev(dev);
3491 	return err;
3492 }
3493 
3494 int mlx4_en_reset_config(struct net_device *dev,
3495 			 struct hwtstamp_config ts_config,
3496 			 netdev_features_t features)
3497 {
3498 	struct mlx4_en_priv *priv = netdev_priv(dev);
3499 	struct mlx4_en_dev *mdev = priv->mdev;
3500 	struct mlx4_en_port_profile new_prof;
3501 	struct mlx4_en_priv *tmp;
3502 	int port_up = 0;
3503 	int err = 0;
3504 
3505 	if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3506 	    priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3507 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3508 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3509 		return 0; /* Nothing to change */
3510 
3511 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3512 	    (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3513 	    (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3514 		en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3515 		return -EINVAL;
3516 	}
3517 
3518 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3519 	if (!tmp)
3520 		return -ENOMEM;
3521 
3522 	mutex_lock(&mdev->state_lock);
3523 
3524 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3525 	memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3526 
3527 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3528 	if (err)
3529 		goto out;
3530 
3531 	if (priv->port_up) {
3532 		port_up = 1;
3533 		mlx4_en_stop_port(dev, 1);
3534 	}
3535 
3536 	mlx4_en_safe_replace_resources(priv, tmp);
3537 
3538 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3539 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3540 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3541 		else
3542 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3543 	} else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3544 		/* RX time-stamping is OFF, update the RX vlan offload
3545 		 * to the latest wanted state
3546 		 */
3547 		if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3548 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3549 		else
3550 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3551 	}
3552 
3553 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3554 		if (features & NETIF_F_RXFCS)
3555 			dev->features |= NETIF_F_RXFCS;
3556 		else
3557 			dev->features &= ~NETIF_F_RXFCS;
3558 	}
3559 
3560 	/* RX vlan offload and RX time-stamping can't co-exist !
3561 	 * Regardless of the caller's choice,
3562 	 * Turn Off RX vlan offload in case of time-stamping is ON
3563 	 */
3564 	if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3565 		if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3566 			en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3567 		dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3568 	}
3569 
3570 	if (port_up) {
3571 		err = mlx4_en_start_port(dev);
3572 		if (err)
3573 			en_err(priv, "Failed starting port\n");
3574 	}
3575 
3576 	if (!err)
3577 		err = mlx4_en_moderation_update(priv);
3578 out:
3579 	mutex_unlock(&mdev->state_lock);
3580 	kfree(tmp);
3581 	if (!err)
3582 		netdev_features_change(dev);
3583 	return err;
3584 }
3585