xref: /linux/drivers/net/ethernet/mellanox/mlx4/en_netdev.c (revision 8f7aa3d3c7323f4ca2768a9e74ebbe359c4f8f88)
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 #include <net/rps.h>
46 #include <net/netdev_queues.h>
47 
48 #include <linux/mlx4/driver.h>
49 #include <linux/mlx4/device.h>
50 #include <linux/mlx4/cmd.h>
51 #include <linux/mlx4/cq.h>
52 
53 #include "mlx4_en.h"
54 #include "en_port.h"
55 
56 #define MLX4_EN_MAX_XDP_MTU ((int)(PAGE_SIZE - ETH_HLEN - (2 * VLAN_HLEN) - \
57 				XDP_PACKET_HEADROOM -			    \
58 				SKB_DATA_ALIGN(sizeof(struct skb_shared_info))))
59 
60 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
61 {
62 	struct mlx4_en_priv *priv = netdev_priv(dev);
63 	int i;
64 	unsigned int offset = 0;
65 
66 	if (up && up != MLX4_EN_NUM_UP_HIGH)
67 		return -EINVAL;
68 
69 	netdev_set_num_tc(dev, up);
70 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
71 	/* Partition Tx queues evenly amongst UP's */
72 	for (i = 0; i < up; i++) {
73 		netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
74 		offset += priv->num_tx_rings_p_up;
75 	}
76 
77 #ifdef CONFIG_MLX4_EN_DCB
78 	if (!mlx4_is_slave(priv->mdev->dev)) {
79 		if (up) {
80 			if (priv->dcbx_cap)
81 				priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
82 		} else {
83 			priv->flags &= ~MLX4_EN_FLAG_DCB_ENABLED;
84 			priv->cee_config.pfc_state = false;
85 		}
86 	}
87 #endif /* CONFIG_MLX4_EN_DCB */
88 
89 	return 0;
90 }
91 
92 int mlx4_en_alloc_tx_queue_per_tc(struct net_device *dev, u8 tc)
93 {
94 	struct mlx4_en_priv *priv = netdev_priv(dev);
95 	struct mlx4_en_dev *mdev = priv->mdev;
96 	struct mlx4_en_port_profile new_prof;
97 	struct mlx4_en_priv *tmp;
98 	int total_count;
99 	int port_up = 0;
100 	int err = 0;
101 
102 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
103 	if (!tmp)
104 		return -ENOMEM;
105 
106 	mutex_lock(&mdev->state_lock);
107 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
108 	new_prof.num_up = (tc == 0) ? MLX4_EN_NUM_UP_LOW :
109 				      MLX4_EN_NUM_UP_HIGH;
110 	new_prof.tx_ring_num[TX] = new_prof.num_tx_rings_p_up *
111 				   new_prof.num_up;
112 	total_count = new_prof.tx_ring_num[TX] + new_prof.tx_ring_num[TX_XDP];
113 	if (total_count > MAX_TX_RINGS) {
114 		err = -EINVAL;
115 		en_err(priv,
116 		       "Total number of TX and XDP rings (%d) exceeds the maximum supported (%d)\n",
117 		       total_count, MAX_TX_RINGS);
118 		goto out;
119 	}
120 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
121 	if (err)
122 		goto out;
123 
124 	if (priv->port_up) {
125 		port_up = 1;
126 		mlx4_en_stop_port(dev, 1);
127 	}
128 
129 	mlx4_en_safe_replace_resources(priv, tmp);
130 	if (port_up) {
131 		err = mlx4_en_start_port(dev);
132 		if (err) {
133 			en_err(priv, "Failed starting port for setup TC\n");
134 			goto out;
135 		}
136 	}
137 
138 	err = mlx4_en_setup_tc(dev, tc);
139 out:
140 	mutex_unlock(&mdev->state_lock);
141 	kfree(tmp);
142 	return err;
143 }
144 
145 static int __mlx4_en_setup_tc(struct net_device *dev, enum tc_setup_type type,
146 			      void *type_data)
147 {
148 	struct tc_mqprio_qopt *mqprio = type_data;
149 
150 	if (type != TC_SETUP_QDISC_MQPRIO)
151 		return -EOPNOTSUPP;
152 
153 	if (mqprio->num_tc && mqprio->num_tc != MLX4_EN_NUM_UP_HIGH)
154 		return -EINVAL;
155 
156 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
157 
158 	return mlx4_en_alloc_tx_queue_per_tc(dev, mqprio->num_tc);
159 }
160 
161 #ifdef CONFIG_RFS_ACCEL
162 
163 struct mlx4_en_filter {
164 	struct list_head next;
165 	struct work_struct work;
166 
167 	u8     ip_proto;
168 	__be32 src_ip;
169 	__be32 dst_ip;
170 	__be16 src_port;
171 	__be16 dst_port;
172 
173 	int rxq_index;
174 	struct mlx4_en_priv *priv;
175 	u32 flow_id;			/* RFS infrastructure id */
176 	int id;				/* mlx4_en driver id */
177 	u64 reg_id;			/* Flow steering API id */
178 	u8 activated;			/* Used to prevent expiry before filter
179 					 * is attached
180 					 */
181 	struct hlist_node filter_chain;
182 };
183 
184 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
185 
186 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
187 {
188 	switch (ip_proto) {
189 	case IPPROTO_UDP:
190 		return MLX4_NET_TRANS_RULE_ID_UDP;
191 	case IPPROTO_TCP:
192 		return MLX4_NET_TRANS_RULE_ID_TCP;
193 	default:
194 		return MLX4_NET_TRANS_RULE_NUM;
195 	}
196 };
197 
198 /* Must not acquire state_lock, as its corresponding work_sync
199  * is done under it.
200  */
201 static void mlx4_en_filter_work(struct work_struct *work)
202 {
203 	struct mlx4_en_filter *filter = container_of(work,
204 						     struct mlx4_en_filter,
205 						     work);
206 	struct mlx4_en_priv *priv = filter->priv;
207 	struct mlx4_spec_list spec_tcp_udp = {
208 		.id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
209 		{
210 			.tcp_udp = {
211 				.dst_port = filter->dst_port,
212 				.dst_port_msk = (__force __be16)-1,
213 				.src_port = filter->src_port,
214 				.src_port_msk = (__force __be16)-1,
215 			},
216 		},
217 	};
218 	struct mlx4_spec_list spec_ip = {
219 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
220 		{
221 			.ipv4 = {
222 				.dst_ip = filter->dst_ip,
223 				.dst_ip_msk = (__force __be32)-1,
224 				.src_ip = filter->src_ip,
225 				.src_ip_msk = (__force __be32)-1,
226 			},
227 		},
228 	};
229 	struct mlx4_spec_list spec_eth = {
230 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
231 	};
232 	struct mlx4_net_trans_rule rule = {
233 		.list = LIST_HEAD_INIT(rule.list),
234 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
235 		.exclusive = 1,
236 		.allow_loopback = 1,
237 		.promisc_mode = MLX4_FS_REGULAR,
238 		.port = priv->port,
239 		.priority = MLX4_DOMAIN_RFS,
240 	};
241 	int rc;
242 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
243 
244 	if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
245 		en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
246 			filter->ip_proto);
247 		goto ignore;
248 	}
249 	list_add_tail(&spec_eth.list, &rule.list);
250 	list_add_tail(&spec_ip.list, &rule.list);
251 	list_add_tail(&spec_tcp_udp.list, &rule.list);
252 
253 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
254 	memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
255 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
256 
257 	filter->activated = 0;
258 
259 	if (filter->reg_id) {
260 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
261 		if (rc && rc != -ENOENT)
262 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
263 	}
264 
265 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
266 	if (rc)
267 		en_err(priv, "Error attaching flow. err = %d\n", rc);
268 
269 ignore:
270 	mlx4_en_filter_rfs_expire(priv);
271 
272 	filter->activated = 1;
273 }
274 
275 static inline struct hlist_head *
276 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
277 		   __be16 src_port, __be16 dst_port)
278 {
279 	unsigned long l;
280 	int bucket_idx;
281 
282 	l = (__force unsigned long)src_port |
283 	    ((__force unsigned long)dst_port << 2);
284 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
285 
286 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
287 
288 	return &priv->filter_hash[bucket_idx];
289 }
290 
291 static struct mlx4_en_filter *
292 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
293 		     __be32 dst_ip, u8 ip_proto, __be16 src_port,
294 		     __be16 dst_port, u32 flow_id)
295 {
296 	struct mlx4_en_filter *filter;
297 
298 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
299 	if (!filter)
300 		return NULL;
301 
302 	filter->priv = priv;
303 	filter->rxq_index = rxq_index;
304 	INIT_WORK(&filter->work, mlx4_en_filter_work);
305 
306 	filter->src_ip = src_ip;
307 	filter->dst_ip = dst_ip;
308 	filter->ip_proto = ip_proto;
309 	filter->src_port = src_port;
310 	filter->dst_port = dst_port;
311 
312 	filter->flow_id = flow_id;
313 
314 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
315 
316 	list_add_tail(&filter->next, &priv->filters);
317 	hlist_add_head(&filter->filter_chain,
318 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
319 					  dst_port));
320 
321 	return filter;
322 }
323 
324 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
325 {
326 	struct mlx4_en_priv *priv = filter->priv;
327 	int rc;
328 
329 	list_del(&filter->next);
330 
331 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
332 	if (rc && rc != -ENOENT)
333 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
334 
335 	kfree(filter);
336 }
337 
338 static inline struct mlx4_en_filter *
339 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
340 		    u8 ip_proto, __be16 src_port, __be16 dst_port)
341 {
342 	struct mlx4_en_filter *filter;
343 	struct mlx4_en_filter *ret = NULL;
344 
345 	hlist_for_each_entry(filter,
346 			     filter_hash_bucket(priv, src_ip, dst_ip,
347 						src_port, dst_port),
348 			     filter_chain) {
349 		if (filter->src_ip == src_ip &&
350 		    filter->dst_ip == dst_ip &&
351 		    filter->ip_proto == ip_proto &&
352 		    filter->src_port == src_port &&
353 		    filter->dst_port == dst_port) {
354 			ret = filter;
355 			break;
356 		}
357 	}
358 
359 	return ret;
360 }
361 
362 static int
363 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
364 		   u16 rxq_index, u32 flow_id)
365 {
366 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
367 	struct mlx4_en_filter *filter;
368 	const struct iphdr *ip;
369 	const __be16 *ports;
370 	u8 ip_proto;
371 	__be32 src_ip;
372 	__be32 dst_ip;
373 	__be16 src_port;
374 	__be16 dst_port;
375 	int nhoff = skb_network_offset(skb);
376 	int ret = 0;
377 
378 	if (skb->encapsulation)
379 		return -EPROTONOSUPPORT;
380 
381 	if (skb->protocol != htons(ETH_P_IP))
382 		return -EPROTONOSUPPORT;
383 
384 	ip = (const struct iphdr *)(skb->data + nhoff);
385 	if (ip_is_fragment(ip))
386 		return -EPROTONOSUPPORT;
387 
388 	if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
389 		return -EPROTONOSUPPORT;
390 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
391 
392 	ip_proto = ip->protocol;
393 	src_ip = ip->saddr;
394 	dst_ip = ip->daddr;
395 	src_port = ports[0];
396 	dst_port = ports[1];
397 
398 	spin_lock_bh(&priv->filters_lock);
399 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
400 				     src_port, dst_port);
401 	if (filter) {
402 		if (filter->rxq_index == rxq_index)
403 			goto out;
404 
405 		filter->rxq_index = rxq_index;
406 	} else {
407 		filter = mlx4_en_filter_alloc(priv, rxq_index,
408 					      src_ip, dst_ip, ip_proto,
409 					      src_port, dst_port, flow_id);
410 		if (!filter) {
411 			ret = -ENOMEM;
412 			goto err;
413 		}
414 	}
415 
416 	queue_work(priv->mdev->workqueue, &filter->work);
417 
418 out:
419 	ret = filter->id;
420 err:
421 	spin_unlock_bh(&priv->filters_lock);
422 
423 	return ret;
424 }
425 
426 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
427 {
428 	struct mlx4_en_filter *filter, *tmp;
429 	LIST_HEAD(del_list);
430 
431 	spin_lock_bh(&priv->filters_lock);
432 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
433 		list_move(&filter->next, &del_list);
434 		hlist_del(&filter->filter_chain);
435 	}
436 	spin_unlock_bh(&priv->filters_lock);
437 
438 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
439 		cancel_work_sync(&filter->work);
440 		mlx4_en_filter_free(filter);
441 	}
442 }
443 
444 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
445 {
446 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
447 	LIST_HEAD(del_list);
448 	int i = 0;
449 
450 	spin_lock_bh(&priv->filters_lock);
451 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
452 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
453 			break;
454 
455 		if (filter->activated &&
456 		    !work_pending(&filter->work) &&
457 		    rps_may_expire_flow(priv->dev,
458 					filter->rxq_index, filter->flow_id,
459 					filter->id)) {
460 			list_move(&filter->next, &del_list);
461 			hlist_del(&filter->filter_chain);
462 		} else
463 			last_filter = filter;
464 
465 		i++;
466 	}
467 
468 	if (last_filter && (&last_filter->next != priv->filters.next))
469 		list_move(&priv->filters, &last_filter->next);
470 
471 	spin_unlock_bh(&priv->filters_lock);
472 
473 	list_for_each_entry_safe(filter, tmp, &del_list, next)
474 		mlx4_en_filter_free(filter);
475 }
476 #endif
477 
478 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
479 				   __be16 proto, u16 vid)
480 {
481 	struct mlx4_en_priv *priv = netdev_priv(dev);
482 	struct mlx4_en_dev *mdev = priv->mdev;
483 	int err;
484 	int idx;
485 
486 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
487 
488 	set_bit(vid, priv->active_vlans);
489 
490 	/* Add VID to port VLAN filter */
491 	mutex_lock(&mdev->state_lock);
492 	if (mdev->device_up && priv->port_up) {
493 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
494 		if (err) {
495 			en_err(priv, "Failed configuring VLAN filter\n");
496 			goto out;
497 		}
498 	}
499 	err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
500 	if (err)
501 		en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
502 
503 out:
504 	mutex_unlock(&mdev->state_lock);
505 	return err;
506 }
507 
508 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
509 				    __be16 proto, u16 vid)
510 {
511 	struct mlx4_en_priv *priv = netdev_priv(dev);
512 	struct mlx4_en_dev *mdev = priv->mdev;
513 	int err = 0;
514 
515 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
516 
517 	clear_bit(vid, priv->active_vlans);
518 
519 	/* Remove VID from port VLAN filter */
520 	mutex_lock(&mdev->state_lock);
521 	mlx4_unregister_vlan(mdev->dev, priv->port, vid);
522 
523 	if (mdev->device_up && priv->port_up) {
524 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
525 		if (err)
526 			en_err(priv, "Failed configuring VLAN filter\n");
527 	}
528 	mutex_unlock(&mdev->state_lock);
529 
530 	return err;
531 }
532 
533 static void mlx4_en_u64_to_mac(struct net_device *dev, u64 src_mac)
534 {
535 	u8 addr[ETH_ALEN];
536 
537 	u64_to_ether_addr(src_mac, addr);
538 	eth_hw_addr_set(dev, addr);
539 }
540 
541 
542 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv,
543 				    const unsigned char *addr,
544 				    int qpn, u64 *reg_id)
545 {
546 	int err;
547 
548 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
549 	    priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
550 		return 0; /* do nothing */
551 
552 	err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
553 				    MLX4_DOMAIN_NIC, reg_id);
554 	if (err) {
555 		en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
556 		return err;
557 	}
558 	en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
559 	return 0;
560 }
561 
562 
563 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
564 				const unsigned char *mac, int *qpn, u64 *reg_id)
565 {
566 	struct mlx4_en_dev *mdev = priv->mdev;
567 	struct mlx4_dev *dev = mdev->dev;
568 	int err;
569 
570 	switch (dev->caps.steering_mode) {
571 	case MLX4_STEERING_MODE_B0: {
572 		struct mlx4_qp qp;
573 		u8 gid[16] = {0};
574 
575 		qp.qpn = *qpn;
576 		memcpy(&gid[10], mac, ETH_ALEN);
577 		gid[5] = priv->port;
578 
579 		err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
580 		break;
581 	}
582 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
583 		struct mlx4_spec_list spec_eth = { {NULL} };
584 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
585 
586 		struct mlx4_net_trans_rule rule = {
587 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
588 			.exclusive = 0,
589 			.allow_loopback = 1,
590 			.promisc_mode = MLX4_FS_REGULAR,
591 			.priority = MLX4_DOMAIN_NIC,
592 		};
593 
594 		rule.port = priv->port;
595 		rule.qpn = *qpn;
596 		INIT_LIST_HEAD(&rule.list);
597 
598 		spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
599 		memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
600 		memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
601 		list_add_tail(&spec_eth.list, &rule.list);
602 
603 		err = mlx4_flow_attach(dev, &rule, reg_id);
604 		break;
605 	}
606 	default:
607 		return -EINVAL;
608 	}
609 	if (err)
610 		en_warn(priv, "Failed Attaching Unicast\n");
611 
612 	return err;
613 }
614 
615 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
616 				     const unsigned char *mac,
617 				     int qpn, u64 reg_id)
618 {
619 	struct mlx4_en_dev *mdev = priv->mdev;
620 	struct mlx4_dev *dev = mdev->dev;
621 
622 	switch (dev->caps.steering_mode) {
623 	case MLX4_STEERING_MODE_B0: {
624 		struct mlx4_qp qp;
625 		u8 gid[16] = {0};
626 
627 		qp.qpn = qpn;
628 		memcpy(&gid[10], mac, ETH_ALEN);
629 		gid[5] = priv->port;
630 
631 		mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
632 		break;
633 	}
634 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
635 		mlx4_flow_detach(dev, reg_id);
636 		break;
637 	}
638 	default:
639 		en_err(priv, "Invalid steering mode.\n");
640 	}
641 }
642 
643 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
644 {
645 	struct mlx4_en_dev *mdev = priv->mdev;
646 	struct mlx4_dev *dev = mdev->dev;
647 	int index = 0;
648 	int err = 0;
649 	int *qpn = &priv->base_qpn;
650 	u64 mac = ether_addr_to_u64(priv->dev->dev_addr);
651 
652 	en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
653 	       priv->dev->dev_addr);
654 	index = mlx4_register_mac(dev, priv->port, mac);
655 	if (index < 0) {
656 		err = index;
657 		en_err(priv, "Failed adding MAC: %pM\n",
658 		       priv->dev->dev_addr);
659 		return err;
660 	}
661 
662 	en_info(priv, "Steering Mode %d\n", dev->caps.steering_mode);
663 
664 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
665 		int base_qpn = mlx4_get_base_qpn(dev, priv->port);
666 		*qpn = base_qpn + index;
667 		return 0;
668 	}
669 
670 	err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP,
671 				    MLX4_RES_USAGE_DRIVER);
672 	en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
673 	if (err) {
674 		en_err(priv, "Failed to reserve qp for mac registration\n");
675 		mlx4_unregister_mac(dev, priv->port, mac);
676 		return err;
677 	}
678 
679 	return 0;
680 }
681 
682 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
683 {
684 	struct mlx4_en_dev *mdev = priv->mdev;
685 	struct mlx4_dev *dev = mdev->dev;
686 	int qpn = priv->base_qpn;
687 
688 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
689 		u64 mac = ether_addr_to_u64(priv->dev->dev_addr);
690 		en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
691 		       priv->dev->dev_addr);
692 		mlx4_unregister_mac(dev, priv->port, mac);
693 	} else {
694 		en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
695 		       priv->port, qpn);
696 		mlx4_qp_release_range(dev, qpn, 1);
697 		priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
698 	}
699 }
700 
701 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
702 			       unsigned char *new_mac, unsigned char *prev_mac)
703 {
704 	struct mlx4_en_dev *mdev = priv->mdev;
705 	struct mlx4_dev *dev = mdev->dev;
706 	int err = 0;
707 	u64 new_mac_u64 = ether_addr_to_u64(new_mac);
708 
709 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
710 		struct hlist_head *bucket;
711 		unsigned int mac_hash;
712 		struct mlx4_mac_entry *entry;
713 		struct hlist_node *tmp;
714 		u64 prev_mac_u64 = ether_addr_to_u64(prev_mac);
715 
716 		bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
717 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
718 			if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
719 				mlx4_en_uc_steer_release(priv, entry->mac,
720 							 qpn, entry->reg_id);
721 				mlx4_unregister_mac(dev, priv->port,
722 						    prev_mac_u64);
723 				hlist_del_rcu(&entry->hlist);
724 				synchronize_rcu();
725 				memcpy(entry->mac, new_mac, ETH_ALEN);
726 				entry->reg_id = 0;
727 				mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
728 				hlist_add_head_rcu(&entry->hlist,
729 						   &priv->mac_hash[mac_hash]);
730 				mlx4_register_mac(dev, priv->port, new_mac_u64);
731 				err = mlx4_en_uc_steer_add(priv, new_mac,
732 							   &qpn,
733 							   &entry->reg_id);
734 				if (err)
735 					return err;
736 				if (priv->tunnel_reg_id) {
737 					mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
738 					priv->tunnel_reg_id = 0;
739 				}
740 				err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
741 							       &priv->tunnel_reg_id);
742 				return err;
743 			}
744 		}
745 		return -EINVAL;
746 	}
747 
748 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
749 }
750 
751 static void mlx4_en_update_user_mac(struct mlx4_en_priv *priv,
752 				    unsigned char new_mac[ETH_ALEN + 2])
753 {
754 	struct mlx4_en_dev *mdev = priv->mdev;
755 	int err;
756 
757 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_USER_MAC_EN))
758 		return;
759 
760 	err = mlx4_SET_PORT_user_mac(mdev->dev, priv->port, new_mac);
761 	if (err)
762 		en_err(priv, "Failed to pass user MAC(%pM) to Firmware for port %d, with error %d\n",
763 		       new_mac, priv->port, err);
764 }
765 
766 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
767 			      unsigned char new_mac[ETH_ALEN + 2])
768 {
769 	int err = 0;
770 
771 	if (priv->port_up) {
772 		/* Remove old MAC and insert the new one */
773 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
774 					  new_mac, priv->current_mac);
775 		if (err)
776 			en_err(priv, "Failed changing HW MAC address\n");
777 	} else
778 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
779 
780 	if (!err)
781 		memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
782 
783 	return err;
784 }
785 
786 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
787 {
788 	struct mlx4_en_priv *priv = netdev_priv(dev);
789 	struct mlx4_en_dev *mdev = priv->mdev;
790 	struct sockaddr *saddr = addr;
791 	unsigned char new_mac[ETH_ALEN + 2];
792 	int err;
793 
794 	if (!is_valid_ether_addr(saddr->sa_data))
795 		return -EADDRNOTAVAIL;
796 
797 	mutex_lock(&mdev->state_lock);
798 	memcpy(new_mac, saddr->sa_data, ETH_ALEN);
799 	err = mlx4_en_do_set_mac(priv, new_mac);
800 	if (err)
801 		goto out;
802 
803 	eth_hw_addr_set(dev, saddr->sa_data);
804 	mlx4_en_update_user_mac(priv, new_mac);
805 out:
806 	mutex_unlock(&mdev->state_lock);
807 
808 	return err;
809 }
810 
811 static void mlx4_en_clear_list(struct net_device *dev)
812 {
813 	struct mlx4_en_priv *priv = netdev_priv(dev);
814 	struct mlx4_en_mc_list *tmp, *mc_to_del;
815 
816 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
817 		list_del(&mc_to_del->list);
818 		kfree(mc_to_del);
819 	}
820 }
821 
822 static void mlx4_en_cache_mclist(struct net_device *dev)
823 {
824 	struct mlx4_en_priv *priv = netdev_priv(dev);
825 	struct netdev_hw_addr *ha;
826 	struct mlx4_en_mc_list *tmp;
827 
828 	mlx4_en_clear_list(dev);
829 	netdev_for_each_mc_addr(ha, dev) {
830 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
831 		if (!tmp) {
832 			mlx4_en_clear_list(dev);
833 			return;
834 		}
835 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
836 		list_add_tail(&tmp->list, &priv->mc_list);
837 	}
838 }
839 
840 static void update_mclist_flags(struct mlx4_en_priv *priv,
841 				struct list_head *dst,
842 				struct list_head *src)
843 {
844 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
845 	bool found;
846 
847 	/* Find all the entries that should be removed from dst,
848 	 * These are the entries that are not found in src
849 	 */
850 	list_for_each_entry(dst_tmp, dst, list) {
851 		found = false;
852 		list_for_each_entry(src_tmp, src, list) {
853 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
854 				found = true;
855 				break;
856 			}
857 		}
858 		if (!found)
859 			dst_tmp->action = MCLIST_REM;
860 	}
861 
862 	/* Add entries that exist in src but not in dst
863 	 * mark them as need to add
864 	 */
865 	list_for_each_entry(src_tmp, src, list) {
866 		found = false;
867 		list_for_each_entry(dst_tmp, dst, list) {
868 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
869 				dst_tmp->action = MCLIST_NONE;
870 				found = true;
871 				break;
872 			}
873 		}
874 		if (!found) {
875 			new_mc = kmemdup(src_tmp,
876 					 sizeof(struct mlx4_en_mc_list),
877 					 GFP_KERNEL);
878 			if (!new_mc)
879 				return;
880 
881 			new_mc->action = MCLIST_ADD;
882 			list_add_tail(&new_mc->list, dst);
883 		}
884 	}
885 }
886 
887 static void mlx4_en_set_rx_mode(struct net_device *dev)
888 {
889 	struct mlx4_en_priv *priv = netdev_priv(dev);
890 
891 	if (!priv->port_up)
892 		return;
893 
894 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
895 }
896 
897 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
898 				     struct mlx4_en_dev *mdev)
899 {
900 	int err = 0;
901 
902 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
903 		if (netif_msg_rx_status(priv))
904 			en_warn(priv, "Entering promiscuous mode\n");
905 		priv->flags |= MLX4_EN_FLAG_PROMISC;
906 
907 		/* Enable promiscouos mode */
908 		switch (mdev->dev->caps.steering_mode) {
909 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
910 			err = mlx4_flow_steer_promisc_add(mdev->dev,
911 							  priv->port,
912 							  priv->base_qpn,
913 							  MLX4_FS_ALL_DEFAULT);
914 			if (err)
915 				en_err(priv, "Failed enabling promiscuous mode\n");
916 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
917 			break;
918 
919 		case MLX4_STEERING_MODE_B0:
920 			err = mlx4_unicast_promisc_add(mdev->dev,
921 						       priv->base_qpn,
922 						       priv->port);
923 			if (err)
924 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
925 
926 			/* Add the default qp number as multicast
927 			 * promisc
928 			 */
929 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
930 				err = mlx4_multicast_promisc_add(mdev->dev,
931 								 priv->base_qpn,
932 								 priv->port);
933 				if (err)
934 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
935 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
936 			}
937 			break;
938 
939 		case MLX4_STEERING_MODE_A0:
940 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
941 						     priv->port,
942 						     priv->base_qpn,
943 						     1);
944 			if (err)
945 				en_err(priv, "Failed enabling promiscuous mode\n");
946 			break;
947 		}
948 
949 		/* Disable port multicast filter (unconditionally) */
950 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
951 					  0, MLX4_MCAST_DISABLE);
952 		if (err)
953 			en_err(priv, "Failed disabling multicast filter\n");
954 	}
955 }
956 
957 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
958 				       struct mlx4_en_dev *mdev)
959 {
960 	int err = 0;
961 
962 	if (netif_msg_rx_status(priv))
963 		en_warn(priv, "Leaving promiscuous mode\n");
964 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
965 
966 	/* Disable promiscouos mode */
967 	switch (mdev->dev->caps.steering_mode) {
968 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
969 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
970 						     priv->port,
971 						     MLX4_FS_ALL_DEFAULT);
972 		if (err)
973 			en_err(priv, "Failed disabling promiscuous mode\n");
974 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
975 		break;
976 
977 	case MLX4_STEERING_MODE_B0:
978 		err = mlx4_unicast_promisc_remove(mdev->dev,
979 						  priv->base_qpn,
980 						  priv->port);
981 		if (err)
982 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
983 		/* Disable Multicast promisc */
984 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
985 			err = mlx4_multicast_promisc_remove(mdev->dev,
986 							    priv->base_qpn,
987 							    priv->port);
988 			if (err)
989 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
990 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
991 		}
992 		break;
993 
994 	case MLX4_STEERING_MODE_A0:
995 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
996 					     priv->port,
997 					     priv->base_qpn, 0);
998 		if (err)
999 			en_err(priv, "Failed disabling promiscuous mode\n");
1000 		break;
1001 	}
1002 }
1003 
1004 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
1005 				 struct net_device *dev,
1006 				 struct mlx4_en_dev *mdev)
1007 {
1008 	struct mlx4_en_mc_list *mclist, *tmp;
1009 	u64 mcast_addr = 0;
1010 	u8 mc_list[16] = {0};
1011 	int err = 0;
1012 
1013 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
1014 	if (dev->flags & IFF_ALLMULTI) {
1015 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1016 					  0, MLX4_MCAST_DISABLE);
1017 		if (err)
1018 			en_err(priv, "Failed disabling multicast filter\n");
1019 
1020 		/* Add the default qp number as multicast promisc */
1021 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1022 			switch (mdev->dev->caps.steering_mode) {
1023 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1024 				err = mlx4_flow_steer_promisc_add(mdev->dev,
1025 								  priv->port,
1026 								  priv->base_qpn,
1027 								  MLX4_FS_MC_DEFAULT);
1028 				break;
1029 
1030 			case MLX4_STEERING_MODE_B0:
1031 				err = mlx4_multicast_promisc_add(mdev->dev,
1032 								 priv->base_qpn,
1033 								 priv->port);
1034 				break;
1035 
1036 			case MLX4_STEERING_MODE_A0:
1037 				break;
1038 			}
1039 			if (err)
1040 				en_err(priv, "Failed entering multicast promisc mode\n");
1041 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1042 		}
1043 	} else {
1044 		/* Disable Multicast promisc */
1045 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1046 			switch (mdev->dev->caps.steering_mode) {
1047 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1048 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
1049 								     priv->port,
1050 								     MLX4_FS_MC_DEFAULT);
1051 				break;
1052 
1053 			case MLX4_STEERING_MODE_B0:
1054 				err = mlx4_multicast_promisc_remove(mdev->dev,
1055 								    priv->base_qpn,
1056 								    priv->port);
1057 				break;
1058 
1059 			case MLX4_STEERING_MODE_A0:
1060 				break;
1061 			}
1062 			if (err)
1063 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
1064 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1065 		}
1066 
1067 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1068 					  0, MLX4_MCAST_DISABLE);
1069 		if (err)
1070 			en_err(priv, "Failed disabling multicast filter\n");
1071 
1072 		/* Flush mcast filter and init it with broadcast address */
1073 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1074 				    1, MLX4_MCAST_CONFIG);
1075 
1076 		/* Update multicast list - we cache all addresses so they won't
1077 		 * change while HW is updated holding the command semaphore
1078 		 */
1079 		netif_addr_lock_bh(dev);
1080 		mlx4_en_cache_mclist(dev);
1081 		netif_addr_unlock_bh(dev);
1082 		list_for_each_entry(mclist, &priv->mc_list, list) {
1083 			mcast_addr = ether_addr_to_u64(mclist->addr);
1084 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1085 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
1086 		}
1087 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1088 					  0, MLX4_MCAST_ENABLE);
1089 		if (err)
1090 			en_err(priv, "Failed enabling multicast filter\n");
1091 
1092 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1093 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1094 			if (mclist->action == MCLIST_REM) {
1095 				/* detach this address and delete from list */
1096 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1097 				mc_list[5] = priv->port;
1098 				err = mlx4_multicast_detach(mdev->dev,
1099 							    priv->rss_map.indir_qp,
1100 							    mc_list,
1101 							    MLX4_PROT_ETH,
1102 							    mclist->reg_id);
1103 				if (err)
1104 					en_err(priv, "Fail to detach multicast address\n");
1105 
1106 				if (mclist->tunnel_reg_id) {
1107 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1108 					if (err)
1109 						en_err(priv, "Failed to detach multicast address\n");
1110 				}
1111 
1112 				/* remove from list */
1113 				list_del(&mclist->list);
1114 				kfree(mclist);
1115 			} else if (mclist->action == MCLIST_ADD) {
1116 				/* attach the address */
1117 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1118 				/* needed for B0 steering support */
1119 				mc_list[5] = priv->port;
1120 				err = mlx4_multicast_attach(mdev->dev,
1121 							    priv->rss_map.indir_qp,
1122 							    mc_list,
1123 							    priv->port, 0,
1124 							    MLX4_PROT_ETH,
1125 							    &mclist->reg_id);
1126 				if (err)
1127 					en_err(priv, "Fail to attach multicast address\n");
1128 
1129 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1130 							       &mclist->tunnel_reg_id);
1131 				if (err)
1132 					en_err(priv, "Failed to attach multicast address\n");
1133 			}
1134 		}
1135 	}
1136 }
1137 
1138 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1139 				 struct net_device *dev,
1140 				 struct mlx4_en_dev *mdev)
1141 {
1142 	struct netdev_hw_addr *ha;
1143 	struct mlx4_mac_entry *entry;
1144 	struct hlist_node *tmp;
1145 	bool found;
1146 	u64 mac;
1147 	int err = 0;
1148 	struct hlist_head *bucket;
1149 	unsigned int i;
1150 	int removed = 0;
1151 	u32 prev_flags;
1152 
1153 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1154 	 * since all modification code is protected by mdev->state_lock
1155 	 */
1156 
1157 	/* find what to remove */
1158 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1159 		bucket = &priv->mac_hash[i];
1160 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1161 			found = false;
1162 			netdev_for_each_uc_addr(ha, dev) {
1163 				if (ether_addr_equal_64bits(entry->mac,
1164 							    ha->addr)) {
1165 					found = true;
1166 					break;
1167 				}
1168 			}
1169 
1170 			/* MAC address of the port is not in uc list */
1171 			if (ether_addr_equal_64bits(entry->mac,
1172 						    priv->current_mac))
1173 				found = true;
1174 
1175 			if (!found) {
1176 				mac = ether_addr_to_u64(entry->mac);
1177 				mlx4_en_uc_steer_release(priv, entry->mac,
1178 							 priv->base_qpn,
1179 							 entry->reg_id);
1180 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1181 
1182 				hlist_del_rcu(&entry->hlist);
1183 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1184 				       entry->mac, priv->port);
1185 				kfree_rcu(entry, rcu);
1186 				++removed;
1187 			}
1188 		}
1189 	}
1190 
1191 	/* if we didn't remove anything, there is no use in trying to add
1192 	 * again once we are in a forced promisc mode state
1193 	 */
1194 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1195 		return;
1196 
1197 	prev_flags = priv->flags;
1198 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1199 
1200 	/* find what to add */
1201 	netdev_for_each_uc_addr(ha, dev) {
1202 		found = false;
1203 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1204 		hlist_for_each_entry(entry, bucket, hlist) {
1205 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1206 				found = true;
1207 				break;
1208 			}
1209 		}
1210 
1211 		if (!found) {
1212 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1213 			if (!entry) {
1214 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1215 				       ha->addr, priv->port);
1216 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1217 				break;
1218 			}
1219 			mac = ether_addr_to_u64(ha->addr);
1220 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1221 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1222 			if (err < 0) {
1223 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1224 				       ha->addr, priv->port, err);
1225 				kfree(entry);
1226 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1227 				break;
1228 			}
1229 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1230 						   &priv->base_qpn,
1231 						   &entry->reg_id);
1232 			if (err) {
1233 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1234 				       ha->addr, priv->port, err);
1235 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1236 				kfree(entry);
1237 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1238 				break;
1239 			} else {
1240 				unsigned int mac_hash;
1241 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1242 				       ha->addr, priv->port);
1243 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1244 				bucket = &priv->mac_hash[mac_hash];
1245 				hlist_add_head_rcu(&entry->hlist, bucket);
1246 			}
1247 		}
1248 	}
1249 
1250 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1251 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1252 			priv->port);
1253 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1254 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1255 			priv->port);
1256 	}
1257 }
1258 
1259 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1260 {
1261 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1262 						 rx_mode_task);
1263 	struct mlx4_en_dev *mdev = priv->mdev;
1264 	struct net_device *dev = priv->dev;
1265 
1266 	mutex_lock(&mdev->state_lock);
1267 	if (!mdev->device_up) {
1268 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1269 		goto out;
1270 	}
1271 	if (!priv->port_up) {
1272 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1273 		goto out;
1274 	}
1275 
1276 	if (!netif_carrier_ok(dev)) {
1277 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1278 			if (priv->port_state.link_state) {
1279 				netif_carrier_on(dev);
1280 				en_dbg(LINK, priv, "Link Up\n");
1281 			}
1282 		}
1283 	}
1284 
1285 	if (dev->priv_flags & IFF_UNICAST_FLT)
1286 		mlx4_en_do_uc_filter(priv, dev, mdev);
1287 
1288 	/* Promsicuous mode: disable all filters */
1289 	if ((dev->flags & IFF_PROMISC) ||
1290 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1291 		mlx4_en_set_promisc_mode(priv, mdev);
1292 		goto out;
1293 	}
1294 
1295 	/* Not in promiscuous mode */
1296 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1297 		mlx4_en_clear_promisc_mode(priv, mdev);
1298 
1299 	mlx4_en_do_multicast(priv, dev, mdev);
1300 out:
1301 	mutex_unlock(&mdev->state_lock);
1302 }
1303 
1304 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1305 {
1306 	u64 reg_id;
1307 	int err = 0;
1308 	int *qpn = &priv->base_qpn;
1309 	struct mlx4_mac_entry *entry;
1310 
1311 	err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1312 	if (err)
1313 		return err;
1314 
1315 	err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1316 				       &priv->tunnel_reg_id);
1317 	if (err)
1318 		goto tunnel_err;
1319 
1320 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1321 	if (!entry) {
1322 		err = -ENOMEM;
1323 		goto alloc_err;
1324 	}
1325 
1326 	memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1327 	memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1328 	entry->reg_id = reg_id;
1329 	hlist_add_head_rcu(&entry->hlist,
1330 			   &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1331 
1332 	return 0;
1333 
1334 alloc_err:
1335 	if (priv->tunnel_reg_id)
1336 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1337 
1338 tunnel_err:
1339 	mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1340 	return err;
1341 }
1342 
1343 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1344 {
1345 	u64 mac;
1346 	unsigned int i;
1347 	int qpn = priv->base_qpn;
1348 	struct hlist_head *bucket;
1349 	struct hlist_node *tmp;
1350 	struct mlx4_mac_entry *entry;
1351 
1352 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1353 		bucket = &priv->mac_hash[i];
1354 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1355 			mac = ether_addr_to_u64(entry->mac);
1356 			en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1357 			       entry->mac);
1358 			mlx4_en_uc_steer_release(priv, entry->mac,
1359 						 qpn, entry->reg_id);
1360 
1361 			mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1362 			hlist_del_rcu(&entry->hlist);
1363 			kfree_rcu(entry, rcu);
1364 		}
1365 	}
1366 
1367 	if (priv->tunnel_reg_id) {
1368 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1369 		priv->tunnel_reg_id = 0;
1370 	}
1371 }
1372 
1373 static void mlx4_en_tx_timeout(struct net_device *dev, unsigned int txqueue)
1374 {
1375 	struct mlx4_en_priv *priv = netdev_priv(dev);
1376 	struct mlx4_en_dev *mdev = priv->mdev;
1377 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][txqueue];
1378 
1379 	if (netif_msg_timer(priv))
1380 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1381 
1382 	en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1383 		txqueue, tx_ring->qpn, tx_ring->sp_cqn,
1384 		tx_ring->cons, tx_ring->prod);
1385 
1386 	priv->port_stats.tx_timeout++;
1387 	if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state)) {
1388 		en_dbg(DRV, priv, "Scheduling port restart\n");
1389 		queue_work(mdev->workqueue, &priv->restart_task);
1390 	}
1391 }
1392 
1393 
1394 static void
1395 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1396 {
1397 	struct mlx4_en_priv *priv = netdev_priv(dev);
1398 
1399 	spin_lock_bh(&priv->stats_lock);
1400 	mlx4_en_fold_software_stats(dev);
1401 	netdev_stats_to_stats64(stats, &dev->stats);
1402 	spin_unlock_bh(&priv->stats_lock);
1403 }
1404 
1405 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1406 {
1407 	struct mlx4_en_cq *cq;
1408 	int i, t;
1409 
1410 	/* If we haven't received a specific coalescing setting
1411 	 * (module param), we set the moderation parameters as follows:
1412 	 * - moder_cnt is set to the number of mtu sized packets to
1413 	 *   satisfy our coalescing target.
1414 	 * - moder_time is set to a fixed value.
1415 	 */
1416 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1417 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1418 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1419 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1420 	en_dbg(INTR, priv, "Default coalescing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1421 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1422 
1423 	/* Setup cq moderation params */
1424 	for (i = 0; i < priv->rx_ring_num; i++) {
1425 		cq = priv->rx_cq[i];
1426 		cq->moder_cnt = priv->rx_frames;
1427 		cq->moder_time = priv->rx_usecs;
1428 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1429 		priv->last_moder_packets[i] = 0;
1430 		priv->last_moder_bytes[i] = 0;
1431 	}
1432 
1433 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1434 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1435 			cq = priv->tx_cq[t][i];
1436 			cq->moder_cnt = priv->tx_frames;
1437 			cq->moder_time = priv->tx_usecs;
1438 		}
1439 	}
1440 
1441 	/* Reset auto-moderation params */
1442 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1443 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1444 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1445 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1446 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1447 	priv->adaptive_rx_coal = 1;
1448 	priv->last_moder_jiffies = 0;
1449 	priv->last_moder_tx_packets = 0;
1450 }
1451 
1452 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1453 {
1454 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1455 	u32 pkt_rate_high, pkt_rate_low;
1456 	struct mlx4_en_cq *cq;
1457 	unsigned long packets;
1458 	unsigned long rate;
1459 	unsigned long avg_pkt_size;
1460 	unsigned long rx_packets;
1461 	unsigned long rx_bytes;
1462 	unsigned long rx_pkt_diff;
1463 	int moder_time;
1464 	int ring, err;
1465 
1466 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1467 		return;
1468 
1469 	pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1470 	pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1471 
1472 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1473 		rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1474 		rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1475 
1476 		rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1477 		packets = rx_pkt_diff;
1478 		rate = packets * HZ / period;
1479 		avg_pkt_size = packets ? (rx_bytes -
1480 				priv->last_moder_bytes[ring]) / packets : 0;
1481 
1482 		/* Apply auto-moderation only when packet rate
1483 		 * exceeds a rate that it matters */
1484 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1485 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1486 			if (rate <= pkt_rate_low)
1487 				moder_time = priv->rx_usecs_low;
1488 			else if (rate >= pkt_rate_high)
1489 				moder_time = priv->rx_usecs_high;
1490 			else
1491 				moder_time = (rate - pkt_rate_low) *
1492 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1493 					(pkt_rate_high - pkt_rate_low) +
1494 					priv->rx_usecs_low;
1495 		} else {
1496 			moder_time = priv->rx_usecs_low;
1497 		}
1498 
1499 		cq = priv->rx_cq[ring];
1500 		if (moder_time != priv->last_moder_time[ring] ||
1501 		    cq->moder_cnt != priv->rx_frames) {
1502 			priv->last_moder_time[ring] = moder_time;
1503 			cq->moder_time = moder_time;
1504 			cq->moder_cnt = priv->rx_frames;
1505 			err = mlx4_en_set_cq_moder(priv, cq);
1506 			if (err)
1507 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1508 				       ring);
1509 		}
1510 		priv->last_moder_packets[ring] = rx_packets;
1511 		priv->last_moder_bytes[ring] = rx_bytes;
1512 	}
1513 
1514 	priv->last_moder_jiffies = jiffies;
1515 }
1516 
1517 static void mlx4_en_do_get_stats(struct work_struct *work)
1518 {
1519 	struct delayed_work *delay = to_delayed_work(work);
1520 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1521 						 stats_task);
1522 	struct mlx4_en_dev *mdev = priv->mdev;
1523 	int err;
1524 
1525 	mutex_lock(&mdev->state_lock);
1526 	if (mdev->device_up) {
1527 		if (priv->port_up) {
1528 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1529 			if (err)
1530 				en_dbg(HW, priv, "Could not update stats\n");
1531 
1532 			mlx4_en_auto_moderation(priv);
1533 		}
1534 
1535 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1536 	}
1537 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1538 		mlx4_en_do_set_mac(priv, priv->current_mac);
1539 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1540 	}
1541 	mutex_unlock(&mdev->state_lock);
1542 }
1543 
1544 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1545  * periodically
1546  */
1547 static void mlx4_en_service_task(struct work_struct *work)
1548 {
1549 	struct delayed_work *delay = to_delayed_work(work);
1550 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1551 						 service_task);
1552 	struct mlx4_en_dev *mdev = priv->mdev;
1553 
1554 	mutex_lock(&mdev->state_lock);
1555 	if (mdev->device_up) {
1556 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1557 			mlx4_en_ptp_overflow_check(mdev);
1558 
1559 		mlx4_en_recover_from_oom(priv);
1560 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1561 				   SERVICE_TASK_DELAY);
1562 	}
1563 	mutex_unlock(&mdev->state_lock);
1564 }
1565 
1566 static void mlx4_en_linkstate(struct mlx4_en_priv *priv)
1567 {
1568 	struct mlx4_en_port_state *port_state = &priv->port_state;
1569 	struct mlx4_en_dev *mdev = priv->mdev;
1570 	struct net_device *dev = priv->dev;
1571 	bool up;
1572 
1573 	if (mlx4_en_QUERY_PORT(mdev, priv->port))
1574 		port_state->link_state = MLX4_PORT_STATE_DEV_EVENT_PORT_DOWN;
1575 
1576 	up = port_state->link_state == MLX4_PORT_STATE_DEV_EVENT_PORT_UP;
1577 	if (up == netif_carrier_ok(dev))
1578 		netif_carrier_event(dev);
1579 	if (!up) {
1580 		en_info(priv, "Link Down\n");
1581 		netif_carrier_off(dev);
1582 	} else {
1583 		en_info(priv, "Link Up\n");
1584 		netif_carrier_on(dev);
1585 	}
1586 }
1587 
1588 static void mlx4_en_linkstate_work(struct work_struct *work)
1589 {
1590 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1591 						 linkstate_task);
1592 	struct mlx4_en_dev *mdev = priv->mdev;
1593 
1594 	mutex_lock(&mdev->state_lock);
1595 	mlx4_en_linkstate(priv);
1596 	mutex_unlock(&mdev->state_lock);
1597 }
1598 
1599 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1600 {
1601 	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1602 	int numa_node = priv->mdev->dev->numa_node;
1603 
1604 	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1605 		return -ENOMEM;
1606 
1607 	cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1608 			ring->affinity_mask);
1609 	return 0;
1610 }
1611 
1612 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1613 {
1614 	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1615 }
1616 
1617 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1618 				      int tx_ring_idx)
1619 {
1620 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1621 	int rr_index = tx_ring_idx;
1622 
1623 	tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1624 	tx_ring->recycle_ring = priv->rx_ring[rr_index];
1625 	en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1626 	       TX_XDP, tx_ring_idx, rr_index);
1627 }
1628 
1629 int mlx4_en_start_port(struct net_device *dev)
1630 {
1631 	struct mlx4_en_priv *priv = netdev_priv(dev);
1632 	struct mlx4_en_dev *mdev = priv->mdev;
1633 	struct mlx4_en_cq *cq;
1634 	struct mlx4_en_tx_ring *tx_ring;
1635 	int rx_index = 0;
1636 	int err = 0;
1637 	int i, t;
1638 	int j;
1639 	u8 mc_list[16] = {0};
1640 
1641 	if (priv->port_up) {
1642 		en_dbg(DRV, priv, "start port called while port already up\n");
1643 		return 0;
1644 	}
1645 
1646 	INIT_LIST_HEAD(&priv->mc_list);
1647 	INIT_LIST_HEAD(&priv->curr_list);
1648 	INIT_LIST_HEAD(&priv->ethtool_list);
1649 	memset(&priv->ethtool_rules[0], 0,
1650 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1651 
1652 	/* Calculate Rx buf size */
1653 	WRITE_ONCE(dev->mtu, min(dev->mtu, priv->max_mtu));
1654 	mlx4_en_calc_rx_buf(dev);
1655 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1656 
1657 	/* Configure rx cq's and rings */
1658 	err = mlx4_en_activate_rx_rings(priv);
1659 	if (err) {
1660 		en_err(priv, "Failed to activate RX rings\n");
1661 		return err;
1662 	}
1663 	for (i = 0; i < priv->rx_ring_num; i++) {
1664 		cq = priv->rx_cq[i];
1665 
1666 		err = mlx4_en_init_affinity_hint(priv, i);
1667 		if (err) {
1668 			en_err(priv, "Failed preparing IRQ affinity hint\n");
1669 			goto cq_err;
1670 		}
1671 
1672 		err = mlx4_en_activate_cq(priv, cq, i);
1673 		if (err) {
1674 			en_err(priv, "Failed activating Rx CQ\n");
1675 			mlx4_en_free_affinity_hint(priv, i);
1676 			goto cq_err;
1677 		}
1678 
1679 		for (j = 0; j < cq->size; j++) {
1680 			struct mlx4_cqe *cqe = NULL;
1681 
1682 			cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1683 			      priv->cqe_factor;
1684 			cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1685 		}
1686 
1687 		err = mlx4_en_set_cq_moder(priv, cq);
1688 		if (err) {
1689 			en_err(priv, "Failed setting cq moderation parameters\n");
1690 			mlx4_en_deactivate_cq(priv, cq);
1691 			mlx4_en_free_affinity_hint(priv, i);
1692 			goto cq_err;
1693 		}
1694 		mlx4_en_arm_cq(priv, cq);
1695 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1696 		++rx_index;
1697 	}
1698 
1699 	/* Set qp number */
1700 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1701 	err = mlx4_en_get_qp(priv);
1702 	if (err) {
1703 		en_err(priv, "Failed getting eth qp\n");
1704 		goto cq_err;
1705 	}
1706 	mdev->mac_removed[priv->port] = 0;
1707 
1708 	priv->counter_index =
1709 			mlx4_get_default_counter_index(mdev->dev, priv->port);
1710 
1711 	err = mlx4_en_config_rss_steer(priv);
1712 	if (err) {
1713 		en_err(priv, "Failed configuring rss steering\n");
1714 		goto mac_err;
1715 	}
1716 
1717 	err = mlx4_en_create_drop_qp(priv);
1718 	if (err)
1719 		goto rss_err;
1720 
1721 	/* Configure tx cq's and rings */
1722 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1723 		u8 num_tx_rings_p_up = t == TX ?
1724 			priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1725 
1726 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1727 			/* Configure cq */
1728 			cq = priv->tx_cq[t][i];
1729 			err = mlx4_en_activate_cq(priv, cq, i);
1730 			if (err) {
1731 				en_err(priv, "Failed allocating Tx CQ\n");
1732 				goto tx_err;
1733 			}
1734 			err = mlx4_en_set_cq_moder(priv, cq);
1735 			if (err) {
1736 				en_err(priv, "Failed setting cq moderation parameters\n");
1737 				mlx4_en_deactivate_cq(priv, cq);
1738 				goto tx_err;
1739 			}
1740 			en_dbg(DRV, priv,
1741 			       "Resetting index of collapsed CQ:%d to -1\n", i);
1742 			cq->buf->wqe_index = cpu_to_be16(0xffff);
1743 
1744 			/* Configure ring */
1745 			tx_ring = priv->tx_ring[t][i];
1746 			err = mlx4_en_activate_tx_ring(priv, tx_ring,
1747 						       cq->mcq.cqn,
1748 						       i / num_tx_rings_p_up);
1749 			if (err) {
1750 				en_err(priv, "Failed allocating Tx ring\n");
1751 				mlx4_en_deactivate_cq(priv, cq);
1752 				goto tx_err;
1753 			}
1754 			clear_bit(MLX4_EN_TX_RING_STATE_RECOVERING, &tx_ring->state);
1755 			if (t != TX_XDP) {
1756 				tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1757 				tx_ring->recycle_ring = NULL;
1758 
1759 				/* Arm CQ for TX completions */
1760 				mlx4_en_arm_cq(priv, cq);
1761 
1762 			} else {
1763 				mlx4_en_init_tx_xdp_ring_descs(priv, tx_ring);
1764 				mlx4_en_init_recycle_ring(priv, i);
1765 				/* XDP TX CQ should never be armed */
1766 			}
1767 
1768 			/* Set initial ownership of all Tx TXBBs to SW (1) */
1769 			for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1770 				*((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1771 		}
1772 	}
1773 
1774 	/* Configure port */
1775 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1776 				    priv->rx_skb_size + ETH_FCS_LEN,
1777 				    priv->prof->tx_pause,
1778 				    priv->prof->tx_ppp,
1779 				    priv->prof->rx_pause,
1780 				    priv->prof->rx_ppp);
1781 	if (err) {
1782 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1783 		       priv->port, err);
1784 		goto tx_err;
1785 	}
1786 
1787 	err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1788 	if (err) {
1789 		en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1790 		       dev->mtu, priv->port, err);
1791 		goto tx_err;
1792 	}
1793 
1794 	/* Set default qp number */
1795 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1796 	if (err) {
1797 		en_err(priv, "Failed setting default qp numbers\n");
1798 		goto tx_err;
1799 	}
1800 
1801 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1802 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1803 		if (err) {
1804 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1805 			       err);
1806 			goto tx_err;
1807 		}
1808 	}
1809 
1810 	/* Init port */
1811 	en_dbg(HW, priv, "Initializing port\n");
1812 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1813 	if (err) {
1814 		en_err(priv, "Failed Initializing port\n");
1815 		goto tx_err;
1816 	}
1817 
1818 	/* Set Unicast and VXLAN steering rules */
1819 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1820 	    mlx4_en_set_rss_steer_rules(priv))
1821 		mlx4_warn(mdev, "Failed setting steering rules\n");
1822 
1823 	/* Attach rx QP to broadcast address */
1824 	eth_broadcast_addr(&mc_list[10]);
1825 	mc_list[5] = priv->port; /* needed for B0 steering support */
1826 	if (mlx4_multicast_attach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1827 				  priv->port, 0, MLX4_PROT_ETH,
1828 				  &priv->broadcast_id))
1829 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1830 
1831 	/* Must redo promiscuous mode setup. */
1832 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1833 
1834 	/* Schedule multicast task to populate multicast list */
1835 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1836 
1837 	if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1838 		udp_tunnel_nic_reset_ntf(dev);
1839 
1840 	priv->port_up = true;
1841 
1842 	/* Process all completions if exist to prevent
1843 	 * the queues freezing if they are full
1844 	 */
1845 	for (i = 0; i < priv->rx_ring_num; i++) {
1846 		local_bh_disable();
1847 		napi_schedule(&priv->rx_cq[i]->napi);
1848 		local_bh_enable();
1849 	}
1850 
1851 	clear_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state);
1852 	netif_tx_start_all_queues(dev);
1853 	netif_device_attach(dev);
1854 
1855 	return 0;
1856 
1857 tx_err:
1858 	if (t == MLX4_EN_NUM_TX_TYPES) {
1859 		t--;
1860 		i = priv->tx_ring_num[t];
1861 	}
1862 	while (t >= 0) {
1863 		while (i--) {
1864 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1865 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1866 		}
1867 		if (!t--)
1868 			break;
1869 		i = priv->tx_ring_num[t];
1870 	}
1871 	mlx4_en_destroy_drop_qp(priv);
1872 rss_err:
1873 	mlx4_en_release_rss_steer(priv);
1874 mac_err:
1875 	mlx4_en_put_qp(priv);
1876 cq_err:
1877 	while (rx_index--) {
1878 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1879 		mlx4_en_free_affinity_hint(priv, rx_index);
1880 	}
1881 	for (i = 0; i < priv->rx_ring_num; i++)
1882 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1883 
1884 	return err; /* need to close devices */
1885 }
1886 
1887 
1888 void mlx4_en_stop_port(struct net_device *dev, int detach)
1889 {
1890 	struct mlx4_en_priv *priv = netdev_priv(dev);
1891 	struct mlx4_en_dev *mdev = priv->mdev;
1892 	struct mlx4_en_mc_list *mclist, *tmp;
1893 	struct ethtool_flow_id *flow, *tmp_flow;
1894 	int i, t;
1895 	u8 mc_list[16] = {0};
1896 
1897 	if (!priv->port_up) {
1898 		en_dbg(DRV, priv, "stop port called while port already down\n");
1899 		return;
1900 	}
1901 
1902 	/* close port*/
1903 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1904 
1905 	/* Synchronize with tx routine */
1906 	netif_tx_lock_bh(dev);
1907 	if (detach)
1908 		netif_device_detach(dev);
1909 	netif_tx_stop_all_queues(dev);
1910 	netif_tx_unlock_bh(dev);
1911 
1912 	netif_tx_disable(dev);
1913 
1914 	spin_lock_bh(&priv->stats_lock);
1915 	mlx4_en_fold_software_stats(dev);
1916 	/* Set port as not active */
1917 	priv->port_up = false;
1918 	spin_unlock_bh(&priv->stats_lock);
1919 
1920 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1921 
1922 	/* Promsicuous mode */
1923 	if (mdev->dev->caps.steering_mode ==
1924 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1925 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1926 				 MLX4_EN_FLAG_MC_PROMISC);
1927 		mlx4_flow_steer_promisc_remove(mdev->dev,
1928 					       priv->port,
1929 					       MLX4_FS_ALL_DEFAULT);
1930 		mlx4_flow_steer_promisc_remove(mdev->dev,
1931 					       priv->port,
1932 					       MLX4_FS_MC_DEFAULT);
1933 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1934 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1935 
1936 		/* Disable promiscouos mode */
1937 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1938 					    priv->port);
1939 
1940 		/* Disable Multicast promisc */
1941 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1942 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1943 						      priv->port);
1944 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1945 		}
1946 	}
1947 
1948 	/* Detach All multicasts */
1949 	eth_broadcast_addr(&mc_list[10]);
1950 	mc_list[5] = priv->port; /* needed for B0 steering support */
1951 	mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1952 			      MLX4_PROT_ETH, priv->broadcast_id);
1953 	list_for_each_entry(mclist, &priv->curr_list, list) {
1954 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1955 		mc_list[5] = priv->port;
1956 		mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp,
1957 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1958 		if (mclist->tunnel_reg_id)
1959 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1960 	}
1961 	mlx4_en_clear_list(dev);
1962 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1963 		list_del(&mclist->list);
1964 		kfree(mclist);
1965 	}
1966 
1967 	/* Flush multicast filter */
1968 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1969 
1970 	/* Remove flow steering rules for the port*/
1971 	if (mdev->dev->caps.steering_mode ==
1972 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1973 		ASSERT_RTNL();
1974 		list_for_each_entry_safe(flow, tmp_flow,
1975 					 &priv->ethtool_list, list) {
1976 			mlx4_flow_detach(mdev->dev, flow->id);
1977 			list_del(&flow->list);
1978 		}
1979 	}
1980 
1981 	mlx4_en_destroy_drop_qp(priv);
1982 
1983 	/* Free TX Rings */
1984 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1985 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1986 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1987 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1988 		}
1989 	}
1990 	msleep(10);
1991 
1992 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1993 		for (i = 0; i < priv->tx_ring_num[t]; i++)
1994 			mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1995 
1996 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1997 		mlx4_en_delete_rss_steer_rules(priv);
1998 
1999 	/* Free RSS qps */
2000 	mlx4_en_release_rss_steer(priv);
2001 
2002 	/* Unregister Mac address for the port */
2003 	mlx4_en_put_qp(priv);
2004 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
2005 		mdev->mac_removed[priv->port] = 1;
2006 
2007 	/* Free RX Rings */
2008 	for (i = 0; i < priv->rx_ring_num; i++) {
2009 		struct mlx4_en_cq *cq = priv->rx_cq[i];
2010 
2011 		napi_synchronize(&cq->napi);
2012 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
2013 		mlx4_en_deactivate_cq(priv, cq);
2014 
2015 		mlx4_en_free_affinity_hint(priv, i);
2016 	}
2017 }
2018 
2019 static void mlx4_en_restart(struct work_struct *work)
2020 {
2021 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2022 						 restart_task);
2023 	struct mlx4_en_dev *mdev = priv->mdev;
2024 	struct net_device *dev = priv->dev;
2025 
2026 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
2027 
2028 	rtnl_lock();
2029 	mutex_lock(&mdev->state_lock);
2030 	if (priv->port_up) {
2031 		mlx4_en_stop_port(dev, 1);
2032 		if (mlx4_en_start_port(dev))
2033 			en_err(priv, "Failed restarting port %d\n", priv->port);
2034 	}
2035 	mutex_unlock(&mdev->state_lock);
2036 	rtnl_unlock();
2037 }
2038 
2039 static void mlx4_en_clear_stats(struct net_device *dev)
2040 {
2041 	struct mlx4_en_priv *priv = netdev_priv(dev);
2042 	struct mlx4_en_dev *mdev = priv->mdev;
2043 	struct mlx4_en_tx_ring **tx_ring;
2044 	int i;
2045 
2046 	if (!mlx4_is_slave(mdev->dev))
2047 		if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
2048 			en_dbg(HW, priv, "Failed dumping statistics\n");
2049 
2050 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
2051 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
2052 	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
2053 	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
2054 	memset(&priv->rx_priority_flowstats, 0,
2055 	       sizeof(priv->rx_priority_flowstats));
2056 	memset(&priv->tx_priority_flowstats, 0,
2057 	       sizeof(priv->tx_priority_flowstats));
2058 	memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
2059 
2060 	tx_ring = priv->tx_ring[TX];
2061 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
2062 		tx_ring[i]->bytes = 0;
2063 		tx_ring[i]->packets = 0;
2064 		tx_ring[i]->tx_csum = 0;
2065 		tx_ring[i]->tx_dropped = 0;
2066 		tx_ring[i]->queue_stopped = 0;
2067 		tx_ring[i]->wake_queue = 0;
2068 		tx_ring[i]->tso_packets = 0;
2069 		tx_ring[i]->xmit_more = 0;
2070 	}
2071 	for (i = 0; i < priv->rx_ring_num; i++) {
2072 		priv->rx_ring[i]->bytes = 0;
2073 		priv->rx_ring[i]->packets = 0;
2074 		priv->rx_ring[i]->csum_ok = 0;
2075 		priv->rx_ring[i]->csum_none = 0;
2076 		priv->rx_ring[i]->csum_complete = 0;
2077 		priv->rx_ring[i]->alloc_fail = 0;
2078 	}
2079 }
2080 
2081 static int mlx4_en_open(struct net_device *dev)
2082 {
2083 	struct mlx4_en_priv *priv = netdev_priv(dev);
2084 	struct mlx4_en_dev *mdev = priv->mdev;
2085 	int err = 0;
2086 
2087 	mutex_lock(&mdev->state_lock);
2088 
2089 	if (!mdev->device_up) {
2090 		en_err(priv, "Cannot open - device down/disabled\n");
2091 		err = -EBUSY;
2092 		goto out;
2093 	}
2094 
2095 	/* Reset HW statistics and SW counters */
2096 	mlx4_en_clear_stats(dev);
2097 
2098 	err = mlx4_en_start_port(dev);
2099 	if (err) {
2100 		en_err(priv, "Failed starting port:%d\n", priv->port);
2101 		goto out;
2102 	}
2103 	mlx4_en_linkstate(priv);
2104 out:
2105 	mutex_unlock(&mdev->state_lock);
2106 	return err;
2107 }
2108 
2109 
2110 static int mlx4_en_close(struct net_device *dev)
2111 {
2112 	struct mlx4_en_priv *priv = netdev_priv(dev);
2113 	struct mlx4_en_dev *mdev = priv->mdev;
2114 
2115 	en_dbg(IFDOWN, priv, "Close port called\n");
2116 
2117 	mutex_lock(&mdev->state_lock);
2118 
2119 	mlx4_en_stop_port(dev, 0);
2120 	netif_carrier_off(dev);
2121 
2122 	mutex_unlock(&mdev->state_lock);
2123 	return 0;
2124 }
2125 
2126 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2127 {
2128 	int i, t;
2129 
2130 #ifdef CONFIG_RFS_ACCEL
2131 	priv->dev->rx_cpu_rmap = NULL;
2132 #endif
2133 
2134 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2135 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2136 			if (priv->tx_ring[t] && priv->tx_ring[t][i])
2137 				mlx4_en_destroy_tx_ring(priv,
2138 							&priv->tx_ring[t][i]);
2139 			if (priv->tx_cq[t] && priv->tx_cq[t][i])
2140 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2141 		}
2142 		kfree(priv->tx_ring[t]);
2143 		kfree(priv->tx_cq[t]);
2144 	}
2145 
2146 	for (i = 0; i < priv->rx_ring_num; i++) {
2147 		if (priv->rx_ring[i])
2148 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2149 				priv->prof->rx_ring_size, priv->stride);
2150 		if (priv->rx_cq[i])
2151 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2152 	}
2153 
2154 }
2155 
2156 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2157 {
2158 	struct mlx4_en_port_profile *prof = priv->prof;
2159 	int i, t;
2160 	int node;
2161 
2162 	/* Create tx Rings */
2163 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2164 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2165 			node = cpu_to_node(i % num_online_cpus());
2166 			if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2167 					      prof->tx_ring_size, i, t, node))
2168 				goto err;
2169 
2170 			if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2171 						   prof->tx_ring_size,
2172 						   TXBB_SIZE, node, i))
2173 				goto err;
2174 		}
2175 	}
2176 
2177 	/* Create rx Rings */
2178 	for (i = 0; i < priv->rx_ring_num; i++) {
2179 		node = cpu_to_node(i % num_online_cpus());
2180 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2181 				      prof->rx_ring_size, i, RX, node))
2182 			goto err;
2183 
2184 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2185 					   prof->rx_ring_size, priv->stride,
2186 					   node, i))
2187 			goto err;
2188 
2189 	}
2190 
2191 #ifdef CONFIG_RFS_ACCEL
2192 	priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2193 #endif
2194 
2195 	return 0;
2196 
2197 err:
2198 	en_err(priv, "Failed to allocate NIC resources\n");
2199 	for (i = 0; i < priv->rx_ring_num; i++) {
2200 		if (priv->rx_ring[i])
2201 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2202 						prof->rx_ring_size,
2203 						priv->stride);
2204 		if (priv->rx_cq[i])
2205 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2206 	}
2207 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2208 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2209 			if (priv->tx_ring[t][i])
2210 				mlx4_en_destroy_tx_ring(priv,
2211 							&priv->tx_ring[t][i]);
2212 			if (priv->tx_cq[t][i])
2213 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2214 		}
2215 	}
2216 	return -ENOMEM;
2217 }
2218 
2219 
2220 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2221 			     struct mlx4_en_priv *src,
2222 			     struct mlx4_en_port_profile *prof)
2223 {
2224 	int t;
2225 
2226 	memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2227 	       sizeof(dst->hwtstamp_config));
2228 	dst->num_tx_rings_p_up = prof->num_tx_rings_p_up;
2229 	dst->rx_ring_num = prof->rx_ring_num;
2230 	dst->flags = prof->flags;
2231 	dst->mdev = src->mdev;
2232 	dst->port = src->port;
2233 	dst->dev = src->dev;
2234 	dst->prof = prof;
2235 	dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2236 					 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2237 
2238 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2239 		dst->tx_ring_num[t] = prof->tx_ring_num[t];
2240 		if (!dst->tx_ring_num[t])
2241 			continue;
2242 
2243 		dst->tx_ring[t] = kcalloc(MAX_TX_RINGS,
2244 					  sizeof(struct mlx4_en_tx_ring *),
2245 					  GFP_KERNEL);
2246 		if (!dst->tx_ring[t])
2247 			goto err_free_tx;
2248 
2249 		dst->tx_cq[t] = kcalloc(MAX_TX_RINGS,
2250 					sizeof(struct mlx4_en_cq *),
2251 					GFP_KERNEL);
2252 		if (!dst->tx_cq[t]) {
2253 			kfree(dst->tx_ring[t]);
2254 			goto err_free_tx;
2255 		}
2256 	}
2257 
2258 	return 0;
2259 
2260 err_free_tx:
2261 	while (t--) {
2262 		kfree(dst->tx_ring[t]);
2263 		kfree(dst->tx_cq[t]);
2264 	}
2265 	return -ENOMEM;
2266 }
2267 
2268 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2269 				struct mlx4_en_priv *src)
2270 {
2271 	int t;
2272 	memcpy(dst->rx_ring, src->rx_ring,
2273 	       sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2274 	memcpy(dst->rx_cq, src->rx_cq,
2275 	       sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2276 	memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2277 	       sizeof(dst->hwtstamp_config));
2278 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2279 		dst->tx_ring_num[t] = src->tx_ring_num[t];
2280 		dst->tx_ring[t] = src->tx_ring[t];
2281 		dst->tx_cq[t] = src->tx_cq[t];
2282 	}
2283 	dst->num_tx_rings_p_up = src->num_tx_rings_p_up;
2284 	dst->rx_ring_num = src->rx_ring_num;
2285 	memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2286 }
2287 
2288 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2289 				struct mlx4_en_priv *tmp,
2290 				struct mlx4_en_port_profile *prof,
2291 				bool carry_xdp_prog)
2292 {
2293 	struct bpf_prog *xdp_prog;
2294 	int i, t, ret;
2295 
2296 	ret = mlx4_en_copy_priv(tmp, priv, prof);
2297 	if (ret) {
2298 		en_warn(priv, "%s: mlx4_en_copy_priv() failed, return\n",
2299 			__func__);
2300 		return ret;
2301 	}
2302 
2303 	if (mlx4_en_alloc_resources(tmp)) {
2304 		en_warn(priv,
2305 			"%s: Resource allocation failed, using previous configuration\n",
2306 			__func__);
2307 		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2308 			kfree(tmp->tx_ring[t]);
2309 			kfree(tmp->tx_cq[t]);
2310 		}
2311 		return -ENOMEM;
2312 	}
2313 
2314 	/* All rx_rings has the same xdp_prog.  Pick the first one. */
2315 	xdp_prog = rcu_dereference_protected(
2316 		priv->rx_ring[0]->xdp_prog,
2317 		lockdep_is_held(&priv->mdev->state_lock));
2318 
2319 	if (xdp_prog && carry_xdp_prog) {
2320 		bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2321 		for (i = 0; i < tmp->rx_ring_num; i++)
2322 			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2323 					   xdp_prog);
2324 	}
2325 
2326 	return 0;
2327 }
2328 
2329 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2330 				    struct mlx4_en_priv *tmp)
2331 {
2332 	mlx4_en_free_resources(priv);
2333 	mlx4_en_update_priv(priv, tmp);
2334 }
2335 
2336 void mlx4_en_destroy_netdev(struct net_device *dev)
2337 {
2338 	struct mlx4_en_priv *priv = netdev_priv(dev);
2339 	struct mlx4_en_dev *mdev = priv->mdev;
2340 
2341 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2342 
2343 	/* Unregister device - this will close the port if it was up */
2344 	if (priv->registered)
2345 		unregister_netdev(dev);
2346 
2347 	if (priv->allocated)
2348 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2349 
2350 	cancel_delayed_work(&priv->stats_task);
2351 	cancel_delayed_work(&priv->service_task);
2352 	/* flush any pending task for this netdev */
2353 	flush_workqueue(mdev->workqueue);
2354 
2355 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2356 		mlx4_en_remove_timestamp(mdev);
2357 
2358 	/* Detach the netdev so tasks would not attempt to access it */
2359 	mutex_lock(&mdev->state_lock);
2360 	mdev->pndev[priv->port] = NULL;
2361 	mdev->upper[priv->port] = NULL;
2362 
2363 #ifdef CONFIG_RFS_ACCEL
2364 	mlx4_en_cleanup_filters(priv);
2365 #endif
2366 
2367 	mlx4_en_free_resources(priv);
2368 	mutex_unlock(&mdev->state_lock);
2369 
2370 	free_netdev(dev);
2371 }
2372 
2373 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2374 {
2375 	struct mlx4_en_priv *priv = netdev_priv(dev);
2376 
2377 	if (mtu > MLX4_EN_MAX_XDP_MTU) {
2378 		en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2379 		       mtu, MLX4_EN_MAX_XDP_MTU);
2380 		return false;
2381 	}
2382 
2383 	return true;
2384 }
2385 
2386 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2387 {
2388 	struct mlx4_en_priv *priv = netdev_priv(dev);
2389 	struct mlx4_en_dev *mdev = priv->mdev;
2390 	int err = 0;
2391 
2392 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2393 		 dev->mtu, new_mtu);
2394 
2395 	if (priv->tx_ring_num[TX_XDP] &&
2396 	    !mlx4_en_check_xdp_mtu(dev, new_mtu))
2397 		return -EOPNOTSUPP;
2398 
2399 	WRITE_ONCE(dev->mtu, new_mtu);
2400 
2401 	if (netif_running(dev)) {
2402 		mutex_lock(&mdev->state_lock);
2403 		if (!mdev->device_up) {
2404 			/* NIC is probably restarting - let restart task reset
2405 			 * the port */
2406 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2407 		} else {
2408 			mlx4_en_stop_port(dev, 1);
2409 			err = mlx4_en_start_port(dev);
2410 			if (err) {
2411 				en_err(priv, "Failed restarting port:%d\n",
2412 					 priv->port);
2413 				if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING,
2414 						      &priv->state))
2415 					queue_work(mdev->workqueue, &priv->restart_task);
2416 			}
2417 		}
2418 		mutex_unlock(&mdev->state_lock);
2419 	}
2420 	return 0;
2421 }
2422 
2423 static int mlx4_en_hwtstamp_set(struct net_device *dev,
2424 				struct kernel_hwtstamp_config *config,
2425 				struct netlink_ext_ack *extack)
2426 {
2427 	struct mlx4_en_priv *priv = netdev_priv(dev);
2428 	struct mlx4_en_dev *mdev = priv->mdev;
2429 
2430 	/* device doesn't support time stamping */
2431 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)) {
2432 		NL_SET_ERR_MSG_MOD(extack,
2433 				   "device doesn't support time stamping");
2434 		return -EINVAL;
2435 	}
2436 
2437 	/* TX HW timestamp */
2438 	switch (config->tx_type) {
2439 	case HWTSTAMP_TX_OFF:
2440 	case HWTSTAMP_TX_ON:
2441 		break;
2442 	default:
2443 		return -ERANGE;
2444 	}
2445 
2446 	/* RX HW timestamp */
2447 	switch (config->rx_filter) {
2448 	case HWTSTAMP_FILTER_NONE:
2449 		break;
2450 	case HWTSTAMP_FILTER_ALL:
2451 	case HWTSTAMP_FILTER_SOME:
2452 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2453 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2454 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2455 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2456 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2457 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2458 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2459 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2460 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2461 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2462 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2463 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2464 	case HWTSTAMP_FILTER_NTP_ALL:
2465 		config->rx_filter = HWTSTAMP_FILTER_ALL;
2466 		break;
2467 	default:
2468 		return -ERANGE;
2469 	}
2470 
2471 	if (mlx4_en_reset_config(dev, config, dev->features)) {
2472 		config->tx_type = HWTSTAMP_TX_OFF;
2473 		config->rx_filter = HWTSTAMP_FILTER_NONE;
2474 	}
2475 
2476 	return 0;
2477 }
2478 
2479 static int mlx4_en_hwtstamp_get(struct net_device *dev,
2480 				struct kernel_hwtstamp_config *config)
2481 {
2482 	struct mlx4_en_priv *priv = netdev_priv(dev);
2483 
2484 	*config = priv->hwtstamp_config;
2485 	return 0;
2486 }
2487 
2488 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2489 					      netdev_features_t features)
2490 {
2491 	struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2492 	struct mlx4_en_dev *mdev = en_priv->mdev;
2493 
2494 	/* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2495 	 * enable/disable make sure S-TAG flag is always in same state as
2496 	 * C-TAG.
2497 	 */
2498 	if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2499 	    !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2500 		features |= NETIF_F_HW_VLAN_STAG_RX;
2501 	else
2502 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2503 
2504 	return features;
2505 }
2506 
2507 static int mlx4_en_set_features(struct net_device *netdev,
2508 		netdev_features_t features)
2509 {
2510 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2511 	bool reset = false;
2512 	int ret = 0;
2513 
2514 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2515 		en_info(priv, "Turn %s RX-FCS\n",
2516 			(features & NETIF_F_RXFCS) ? "ON" : "OFF");
2517 		reset = true;
2518 	}
2519 
2520 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2521 		u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2522 
2523 		en_info(priv, "Turn %s RX-ALL\n",
2524 			ignore_fcs_value ? "ON" : "OFF");
2525 		ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2526 					      priv->port, ignore_fcs_value);
2527 		if (ret)
2528 			return ret;
2529 	}
2530 
2531 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2532 		en_info(priv, "Turn %s RX vlan strip offload\n",
2533 			(features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2534 		reset = true;
2535 	}
2536 
2537 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2538 		en_info(priv, "Turn %s TX vlan strip offload\n",
2539 			(features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2540 
2541 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2542 		en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2543 			(features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2544 
2545 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2546 		en_info(priv, "Turn %s loopback\n",
2547 			(features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2548 		mlx4_en_update_loopback_state(netdev, features);
2549 	}
2550 
2551 	if (reset) {
2552 		ret = mlx4_en_reset_config(netdev, &priv->hwtstamp_config,
2553 					   features);
2554 		if (ret)
2555 			return ret;
2556 	}
2557 
2558 	return 0;
2559 }
2560 
2561 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2562 {
2563 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2564 	struct mlx4_en_dev *mdev = en_priv->mdev;
2565 
2566 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac);
2567 }
2568 
2569 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2570 			       __be16 vlan_proto)
2571 {
2572 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2573 	struct mlx4_en_dev *mdev = en_priv->mdev;
2574 
2575 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2576 				vlan_proto);
2577 }
2578 
2579 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2580 			       int max_tx_rate)
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_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2586 				max_tx_rate);
2587 }
2588 
2589 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2590 {
2591 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2592 	struct mlx4_en_dev *mdev = en_priv->mdev;
2593 
2594 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2595 }
2596 
2597 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2598 {
2599 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2600 	struct mlx4_en_dev *mdev = en_priv->mdev;
2601 
2602 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2603 }
2604 
2605 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2606 {
2607 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2608 	struct mlx4_en_dev *mdev = en_priv->mdev;
2609 
2610 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2611 }
2612 
2613 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2614 				struct ifla_vf_stats *vf_stats)
2615 {
2616 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2617 	struct mlx4_en_dev *mdev = en_priv->mdev;
2618 
2619 	return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2620 }
2621 
2622 #define PORT_ID_BYTE_LEN 8
2623 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2624 				    struct netdev_phys_item_id *ppid)
2625 {
2626 	struct mlx4_en_priv *priv = netdev_priv(dev);
2627 	struct mlx4_dev *mdev = priv->mdev->dev;
2628 	int i;
2629 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2630 
2631 	if (!phys_port_id)
2632 		return -EOPNOTSUPP;
2633 
2634 	ppid->id_len = sizeof(phys_port_id);
2635 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2636 		ppid->id[i] =  phys_port_id & 0xff;
2637 		phys_port_id >>= 8;
2638 	}
2639 	return 0;
2640 }
2641 
2642 static int mlx4_udp_tunnel_sync(struct net_device *dev, unsigned int table)
2643 {
2644 	struct mlx4_en_priv *priv = netdev_priv(dev);
2645 	struct udp_tunnel_info ti;
2646 	int ret;
2647 
2648 	udp_tunnel_nic_get_port(dev, table, 0, &ti);
2649 	priv->vxlan_port = ti.port;
2650 
2651 	ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2652 	if (ret)
2653 		return ret;
2654 
2655 	return mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2656 				   VXLAN_STEER_BY_OUTER_MAC,
2657 				   !!priv->vxlan_port);
2658 }
2659 
2660 static const struct udp_tunnel_nic_info mlx4_udp_tunnels = {
2661 	.sync_table	= mlx4_udp_tunnel_sync,
2662 	.flags		= UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
2663 	.tables		= {
2664 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
2665 	},
2666 };
2667 
2668 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2669 						struct net_device *dev,
2670 						netdev_features_t features)
2671 {
2672 	features = vlan_features_check(skb, features);
2673 	features = vxlan_features_check(skb, features);
2674 
2675 	/* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2676 	 * support inner IPv6 checksums and segmentation so  we need to
2677 	 * strip that feature if this is an IPv6 encapsulated frame.
2678 	 */
2679 	if (skb->encapsulation &&
2680 	    (skb->ip_summed == CHECKSUM_PARTIAL)) {
2681 		struct mlx4_en_priv *priv = netdev_priv(dev);
2682 
2683 		if (!priv->vxlan_port ||
2684 		    (ip_hdr(skb)->version != 4) ||
2685 		    (udp_hdr(skb)->dest != priv->vxlan_port))
2686 			features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2687 	}
2688 
2689 	return features;
2690 }
2691 
2692 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2693 {
2694 	struct mlx4_en_priv *priv = netdev_priv(dev);
2695 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2696 	struct mlx4_update_qp_params params;
2697 	int err;
2698 
2699 	if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2700 		return -EOPNOTSUPP;
2701 
2702 	/* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2703 	if (maxrate >> 12) {
2704 		params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2705 		params.rate_val  = maxrate / 1000;
2706 	} else if (maxrate) {
2707 		params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2708 		params.rate_val  = maxrate;
2709 	} else { /* zero serves to revoke the QP rate-limitation */
2710 		params.rate_unit = 0;
2711 		params.rate_val  = 0;
2712 	}
2713 
2714 	err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2715 			     &params);
2716 	return err;
2717 }
2718 
2719 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2720 {
2721 	struct mlx4_en_priv *priv = netdev_priv(dev);
2722 	struct mlx4_en_dev *mdev = priv->mdev;
2723 	struct mlx4_en_port_profile new_prof;
2724 	struct bpf_prog *old_prog;
2725 	struct mlx4_en_priv *tmp;
2726 	int tx_changed = 0;
2727 	int xdp_ring_num;
2728 	int port_up = 0;
2729 	int err;
2730 	int i;
2731 
2732 	xdp_ring_num = prog ? priv->rx_ring_num : 0;
2733 
2734 	/* No need to reconfigure buffers when simply swapping the
2735 	 * program for a new one.
2736 	 */
2737 	if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2738 		if (prog)
2739 			bpf_prog_add(prog, priv->rx_ring_num - 1);
2740 
2741 		mutex_lock(&mdev->state_lock);
2742 		for (i = 0; i < priv->rx_ring_num; i++) {
2743 			old_prog = rcu_dereference_protected(
2744 					priv->rx_ring[i]->xdp_prog,
2745 					lockdep_is_held(&mdev->state_lock));
2746 			rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2747 			if (old_prog)
2748 				bpf_prog_put(old_prog);
2749 		}
2750 		mutex_unlock(&mdev->state_lock);
2751 		return 0;
2752 	}
2753 
2754 	if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2755 		return -EOPNOTSUPP;
2756 
2757 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2758 	if (!tmp)
2759 		return -ENOMEM;
2760 
2761 	if (prog)
2762 		bpf_prog_add(prog, priv->rx_ring_num - 1);
2763 
2764 	mutex_lock(&mdev->state_lock);
2765 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2766 	new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2767 
2768 	if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2769 		tx_changed = 1;
2770 		new_prof.tx_ring_num[TX] =
2771 			MAX_TX_RINGS - ALIGN(xdp_ring_num, priv->prof->num_up);
2772 		en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2773 	}
2774 
2775 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2776 	if (err) {
2777 		if (prog)
2778 			bpf_prog_sub(prog, priv->rx_ring_num - 1);
2779 		goto unlock_out;
2780 	}
2781 
2782 	if (priv->port_up) {
2783 		port_up = 1;
2784 		mlx4_en_stop_port(dev, 1);
2785 	}
2786 
2787 	mlx4_en_safe_replace_resources(priv, tmp);
2788 	if (tx_changed)
2789 		netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2790 
2791 	for (i = 0; i < priv->rx_ring_num; i++) {
2792 		old_prog = rcu_dereference_protected(
2793 					priv->rx_ring[i]->xdp_prog,
2794 					lockdep_is_held(&mdev->state_lock));
2795 		rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2796 		if (old_prog)
2797 			bpf_prog_put(old_prog);
2798 	}
2799 
2800 	if (port_up) {
2801 		err = mlx4_en_start_port(dev);
2802 		if (err) {
2803 			en_err(priv, "Failed starting port %d for XDP change\n",
2804 			       priv->port);
2805 			if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state))
2806 				queue_work(mdev->workqueue, &priv->restart_task);
2807 		}
2808 	}
2809 
2810 unlock_out:
2811 	mutex_unlock(&mdev->state_lock);
2812 	kfree(tmp);
2813 	return err;
2814 }
2815 
2816 static int mlx4_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2817 {
2818 	switch (xdp->command) {
2819 	case XDP_SETUP_PROG:
2820 		return mlx4_xdp_set(dev, xdp->prog);
2821 	default:
2822 		return -EINVAL;
2823 	}
2824 }
2825 
2826 static const struct net_device_ops mlx4_netdev_ops = {
2827 	.ndo_open		= mlx4_en_open,
2828 	.ndo_stop		= mlx4_en_close,
2829 	.ndo_start_xmit		= mlx4_en_xmit,
2830 	.ndo_select_queue	= mlx4_en_select_queue,
2831 	.ndo_get_stats64	= mlx4_en_get_stats64,
2832 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2833 	.ndo_set_mac_address	= mlx4_en_set_mac,
2834 	.ndo_validate_addr	= eth_validate_addr,
2835 	.ndo_change_mtu		= mlx4_en_change_mtu,
2836 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2837 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2838 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2839 	.ndo_set_features	= mlx4_en_set_features,
2840 	.ndo_fix_features	= mlx4_en_fix_features,
2841 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2842 #ifdef CONFIG_RFS_ACCEL
2843 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2844 #endif
2845 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2846 	.ndo_features_check	= mlx4_en_features_check,
2847 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2848 	.ndo_bpf		= mlx4_xdp,
2849 	.ndo_hwtstamp_get	= mlx4_en_hwtstamp_get,
2850 	.ndo_hwtstamp_set	= mlx4_en_hwtstamp_set,
2851 };
2852 
2853 static const struct net_device_ops mlx4_netdev_ops_master = {
2854 	.ndo_open		= mlx4_en_open,
2855 	.ndo_stop		= mlx4_en_close,
2856 	.ndo_start_xmit		= mlx4_en_xmit,
2857 	.ndo_select_queue	= mlx4_en_select_queue,
2858 	.ndo_get_stats64	= mlx4_en_get_stats64,
2859 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2860 	.ndo_set_mac_address	= mlx4_en_set_mac,
2861 	.ndo_validate_addr	= eth_validate_addr,
2862 	.ndo_change_mtu		= mlx4_en_change_mtu,
2863 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2864 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2865 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2866 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2867 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2868 	.ndo_set_vf_rate	= mlx4_en_set_vf_rate,
2869 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2870 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2871 	.ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2872 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2873 	.ndo_set_features	= mlx4_en_set_features,
2874 	.ndo_fix_features	= mlx4_en_fix_features,
2875 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2876 #ifdef CONFIG_RFS_ACCEL
2877 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2878 #endif
2879 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2880 	.ndo_features_check	= mlx4_en_features_check,
2881 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2882 	.ndo_bpf		= mlx4_xdp,
2883 };
2884 
2885 static const struct xdp_metadata_ops mlx4_xdp_metadata_ops = {
2886 	.xmo_rx_timestamp		= mlx4_en_xdp_rx_timestamp,
2887 	.xmo_rx_hash			= mlx4_en_xdp_rx_hash,
2888 };
2889 
2890 int mlx4_en_netdev_event(struct notifier_block *this,
2891 			 unsigned long event, void *ptr)
2892 {
2893 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2894 	u8 port = 0;
2895 	struct mlx4_en_dev *mdev;
2896 	struct mlx4_dev *dev;
2897 	int i, num_eth_ports = 0;
2898 	bool do_bond = true;
2899 	u8 v2p_port1 = 0;
2900 	u8 v2p_port2 = 0;
2901 
2902 	if (!net_eq(dev_net(ndev), &init_net))
2903 		return NOTIFY_DONE;
2904 
2905 	mdev = container_of(this, struct mlx4_en_dev, netdev_nb);
2906 	dev = mdev->dev;
2907 
2908 	/* Go into this mode only when two network devices set on two ports
2909 	 * of the same mlx4 device are slaves of the same bonding master
2910 	 */
2911 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2912 		++num_eth_ports;
2913 		if (!port && (mdev->pndev[i] == ndev))
2914 			port = i;
2915 		mdev->upper[i] = mdev->pndev[i] ?
2916 			netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2917 		/* condition not met: network device is a slave */
2918 		if (!mdev->upper[i])
2919 			do_bond = false;
2920 		if (num_eth_ports < 2)
2921 			continue;
2922 		/* condition not met: same master */
2923 		if (mdev->upper[i] != mdev->upper[i-1])
2924 			do_bond = false;
2925 	}
2926 	/* condition not met: 2 salves */
2927 	do_bond = (num_eth_ports ==  2) ? do_bond : false;
2928 
2929 	/* handle only events that come with enough info */
2930 	if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
2931 		return NOTIFY_DONE;
2932 
2933 	if (do_bond) {
2934 		struct netdev_notifier_bonding_info *notifier_info = ptr;
2935 		struct netdev_bonding_info *bonding_info =
2936 			&notifier_info->bonding_info;
2937 
2938 		/* required mode 1, 2 or 4 */
2939 		if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
2940 		    (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
2941 		    (bonding_info->master.bond_mode != BOND_MODE_8023AD))
2942 			do_bond = false;
2943 
2944 		/* require exactly 2 slaves */
2945 		if (bonding_info->master.num_slaves != 2)
2946 			do_bond = false;
2947 
2948 		/* calc v2p */
2949 		if (do_bond) {
2950 			if (bonding_info->master.bond_mode ==
2951 			    BOND_MODE_ACTIVEBACKUP) {
2952 				/* in active-backup mode virtual ports are
2953 				 * mapped to the physical port of the active
2954 				 * slave */
2955 				if (bonding_info->slave.state ==
2956 				    BOND_STATE_BACKUP) {
2957 					if (port == 1) {
2958 						v2p_port1 = 2;
2959 						v2p_port2 = 2;
2960 					} else {
2961 						v2p_port1 = 1;
2962 						v2p_port2 = 1;
2963 					}
2964 				} else { /* BOND_STATE_ACTIVE */
2965 					if (port == 1) {
2966 						v2p_port1 = 1;
2967 						v2p_port2 = 1;
2968 					} else {
2969 						v2p_port1 = 2;
2970 						v2p_port2 = 2;
2971 					}
2972 				}
2973 			} else { /* Active-Active */
2974 				/* in active-active mode a virtual port is
2975 				 * mapped to the native physical port if and only
2976 				 * if the physical port is up */
2977 				__s8 link = bonding_info->slave.link;
2978 
2979 				if (port == 1)
2980 					v2p_port2 = 2;
2981 				else
2982 					v2p_port1 = 1;
2983 				if ((link == BOND_LINK_UP) ||
2984 				    (link == BOND_LINK_FAIL)) {
2985 					if (port == 1)
2986 						v2p_port1 = 1;
2987 					else
2988 						v2p_port2 = 2;
2989 				} else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
2990 					if (port == 1)
2991 						v2p_port1 = 2;
2992 					else
2993 						v2p_port2 = 1;
2994 				}
2995 			}
2996 		}
2997 	}
2998 
2999 	mlx4_queue_bond_work(dev, do_bond, v2p_port1, v2p_port2);
3000 
3001 	return NOTIFY_DONE;
3002 }
3003 
3004 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3005 				     struct mlx4_en_stats_bitmap *stats_bitmap,
3006 				     u8 rx_ppp, u8 rx_pause,
3007 				     u8 tx_ppp, u8 tx_pause)
3008 {
3009 	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3010 
3011 	if (!mlx4_is_slave(dev) &&
3012 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3013 		mutex_lock(&stats_bitmap->mutex);
3014 		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3015 
3016 		if (rx_ppp)
3017 			bitmap_set(stats_bitmap->bitmap, last_i,
3018 				   NUM_FLOW_PRIORITY_STATS_RX);
3019 		last_i += NUM_FLOW_PRIORITY_STATS_RX;
3020 
3021 		if (rx_pause && !(rx_ppp))
3022 			bitmap_set(stats_bitmap->bitmap, last_i,
3023 				   NUM_FLOW_STATS_RX);
3024 		last_i += NUM_FLOW_STATS_RX;
3025 
3026 		if (tx_ppp)
3027 			bitmap_set(stats_bitmap->bitmap, last_i,
3028 				   NUM_FLOW_PRIORITY_STATS_TX);
3029 		last_i += NUM_FLOW_PRIORITY_STATS_TX;
3030 
3031 		if (tx_pause && !(tx_ppp))
3032 			bitmap_set(stats_bitmap->bitmap, last_i,
3033 				   NUM_FLOW_STATS_TX);
3034 		last_i += NUM_FLOW_STATS_TX;
3035 
3036 		mutex_unlock(&stats_bitmap->mutex);
3037 	}
3038 }
3039 
3040 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3041 			      struct mlx4_en_stats_bitmap *stats_bitmap,
3042 			      u8 rx_ppp, u8 rx_pause,
3043 			      u8 tx_ppp, u8 tx_pause)
3044 {
3045 	int last_i = 0;
3046 
3047 	mutex_init(&stats_bitmap->mutex);
3048 	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3049 
3050 	if (mlx4_is_slave(dev)) {
3051 		bitmap_set(stats_bitmap->bitmap, last_i +
3052 					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3053 		bitmap_set(stats_bitmap->bitmap, last_i +
3054 					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3055 		bitmap_set(stats_bitmap->bitmap, last_i +
3056 					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3057 		bitmap_set(stats_bitmap->bitmap, last_i +
3058 					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3059 		bitmap_set(stats_bitmap->bitmap, last_i +
3060 					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3061 		bitmap_set(stats_bitmap->bitmap, last_i +
3062 					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3063 	} else {
3064 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3065 	}
3066 	last_i += NUM_MAIN_STATS;
3067 
3068 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3069 	last_i += NUM_PORT_STATS;
3070 
3071 	if (mlx4_is_master(dev))
3072 		bitmap_set(stats_bitmap->bitmap, last_i,
3073 			   NUM_PF_STATS);
3074 	last_i += NUM_PF_STATS;
3075 
3076 	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3077 					rx_ppp, rx_pause,
3078 					tx_ppp, tx_pause);
3079 	last_i += NUM_FLOW_STATS;
3080 
3081 	if (!mlx4_is_slave(dev))
3082 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3083 	last_i += NUM_PKT_STATS;
3084 
3085 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3086 	last_i += NUM_XDP_STATS;
3087 
3088 	if (!mlx4_is_slave(dev))
3089 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PHY_STATS);
3090 	last_i += NUM_PHY_STATS;
3091 }
3092 
3093 static void mlx4_get_queue_stats_rx(struct net_device *dev, int i,
3094 				    struct netdev_queue_stats_rx *stats)
3095 {
3096 	struct mlx4_en_priv *priv = netdev_priv(dev);
3097 	const struct mlx4_en_rx_ring *ring;
3098 
3099 	spin_lock_bh(&priv->stats_lock);
3100 
3101 	if (!priv->port_up || mlx4_is_master(priv->mdev->dev))
3102 		goto out_unlock;
3103 
3104 	ring = priv->rx_ring[i];
3105 	stats->packets = READ_ONCE(ring->packets);
3106 	stats->bytes   = READ_ONCE(ring->bytes);
3107 	stats->alloc_fail = READ_ONCE(ring->alloc_fail);
3108 
3109 out_unlock:
3110 	spin_unlock_bh(&priv->stats_lock);
3111 }
3112 
3113 static void mlx4_get_queue_stats_tx(struct net_device *dev, int i,
3114 				    struct netdev_queue_stats_tx *stats)
3115 {
3116 	struct mlx4_en_priv *priv = netdev_priv(dev);
3117 	const struct mlx4_en_tx_ring *ring;
3118 
3119 	spin_lock_bh(&priv->stats_lock);
3120 
3121 	if (!priv->port_up || mlx4_is_master(priv->mdev->dev))
3122 		goto out_unlock;
3123 
3124 	ring = priv->tx_ring[TX][i];
3125 	stats->packets = READ_ONCE(ring->packets);
3126 	stats->bytes   = READ_ONCE(ring->bytes);
3127 
3128 out_unlock:
3129 	spin_unlock_bh(&priv->stats_lock);
3130 }
3131 
3132 static void mlx4_get_base_stats(struct net_device *dev,
3133 				struct netdev_queue_stats_rx *rx,
3134 				struct netdev_queue_stats_tx *tx)
3135 {
3136 	struct mlx4_en_priv *priv = netdev_priv(dev);
3137 
3138 	spin_lock_bh(&priv->stats_lock);
3139 
3140 	if (!priv->port_up || mlx4_is_master(priv->mdev->dev))
3141 		goto out_unlock;
3142 
3143 	if (priv->rx_ring_num) {
3144 		rx->packets = 0;
3145 		rx->bytes = 0;
3146 		rx->alloc_fail = 0;
3147 	}
3148 
3149 	if (priv->tx_ring_num[TX]) {
3150 		tx->packets = 0;
3151 		tx->bytes = 0;
3152 	}
3153 
3154 out_unlock:
3155 	spin_unlock_bh(&priv->stats_lock);
3156 }
3157 
3158 static const struct netdev_stat_ops mlx4_stat_ops = {
3159 	.get_queue_stats_rx     = mlx4_get_queue_stats_rx,
3160 	.get_queue_stats_tx     = mlx4_get_queue_stats_tx,
3161 	.get_base_stats         = mlx4_get_base_stats,
3162 };
3163 
3164 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3165 			struct mlx4_en_port_profile *prof)
3166 {
3167 	struct net_device *dev;
3168 	struct mlx4_en_priv *priv;
3169 	int i, t;
3170 	int err;
3171 
3172 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3173 				 MAX_TX_RINGS, MAX_RX_RINGS);
3174 	if (dev == NULL)
3175 		return -ENOMEM;
3176 
3177 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3178 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3179 
3180 	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3181 	dev->dev_port = port - 1;
3182 
3183 	/*
3184 	 * Initialize driver private data
3185 	 */
3186 
3187 	priv = netdev_priv(dev);
3188 	memset(priv, 0, sizeof(struct mlx4_en_priv));
3189 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3190 	spin_lock_init(&priv->stats_lock);
3191 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3192 	INIT_WORK(&priv->restart_task, mlx4_en_restart);
3193 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate_work);
3194 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3195 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3196 #ifdef CONFIG_RFS_ACCEL
3197 	INIT_LIST_HEAD(&priv->filters);
3198 	spin_lock_init(&priv->filters_lock);
3199 #endif
3200 
3201 	priv->dev = dev;
3202 	priv->mdev = mdev;
3203 	priv->ddev = &mdev->pdev->dev;
3204 	priv->prof = prof;
3205 	priv->port = port;
3206 	priv->port_up = false;
3207 	priv->flags = prof->flags;
3208 	priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3209 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3210 			MLX4_WQE_CTRL_SOLICITED);
3211 	priv->num_tx_rings_p_up = mdev->profile.max_num_tx_rings_p_up;
3212 	priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3213 	netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3214 
3215 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3216 		priv->tx_ring_num[t] = prof->tx_ring_num[t];
3217 		if (!priv->tx_ring_num[t])
3218 			continue;
3219 
3220 		priv->tx_ring[t] = kcalloc(MAX_TX_RINGS,
3221 					   sizeof(struct mlx4_en_tx_ring *),
3222 					   GFP_KERNEL);
3223 		if (!priv->tx_ring[t]) {
3224 			err = -ENOMEM;
3225 			goto out;
3226 		}
3227 		priv->tx_cq[t] = kcalloc(MAX_TX_RINGS,
3228 					 sizeof(struct mlx4_en_cq *),
3229 					 GFP_KERNEL);
3230 		if (!priv->tx_cq[t]) {
3231 			err = -ENOMEM;
3232 			goto out;
3233 		}
3234 	}
3235 	priv->rx_ring_num = prof->rx_ring_num;
3236 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3237 	priv->cqe_size = mdev->dev->caps.cqe_size;
3238 	priv->mac_index = -1;
3239 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
3240 #ifdef CONFIG_MLX4_EN_DCB
3241 	if (!mlx4_is_slave(priv->mdev->dev)) {
3242 		u8 prio;
3243 
3244 		for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; ++prio) {
3245 			priv->ets.prio_tc[prio] = prio;
3246 			priv->ets.tc_tsa[prio]  = IEEE_8021QAZ_TSA_VENDOR;
3247 		}
3248 
3249 		priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3250 			DCB_CAP_DCBX_VER_IEEE;
3251 		priv->flags |= MLX4_EN_DCB_ENABLED;
3252 		priv->cee_config.pfc_state = false;
3253 
3254 		for (i = 0; i < MLX4_EN_NUM_UP_HIGH; i++)
3255 			priv->cee_config.dcb_pfc[i] = pfc_disabled;
3256 
3257 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3258 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3259 		} else {
3260 			en_info(priv, "enabling only PFC DCB ops\n");
3261 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3262 		}
3263 	}
3264 #endif
3265 
3266 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3267 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
3268 
3269 	/* Query for default mac and max mtu */
3270 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3271 
3272 	if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3273 	    MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3274 		priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3275 
3276 	/* Set default MAC */
3277 	dev->addr_len = ETH_ALEN;
3278 	mlx4_en_u64_to_mac(dev, mdev->dev->caps.def_mac[priv->port]);
3279 	if (!is_valid_ether_addr(dev->dev_addr)) {
3280 		en_err(priv, "Port: %d, invalid mac burned: %pM, quitting\n",
3281 		       priv->port, dev->dev_addr);
3282 		err = -EINVAL;
3283 		goto out;
3284 	} else if (mlx4_is_slave(priv->mdev->dev) &&
3285 		   (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3286 		/* Random MAC was assigned in mlx4_slave_cap
3287 		 * in mlx4_core module
3288 		 */
3289 		dev->addr_assign_type |= NET_ADDR_RANDOM;
3290 		en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3291 	}
3292 
3293 	memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3294 
3295 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3296 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3297 	err = mlx4_en_alloc_resources(priv);
3298 	if (err)
3299 		goto out;
3300 
3301 	/* Initialize time stamping config */
3302 	priv->hwtstamp_config.flags = 0;
3303 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3304 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3305 
3306 	/* Allocate page for receive rings */
3307 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3308 				MLX4_EN_PAGE_SIZE);
3309 	if (err) {
3310 		en_err(priv, "Failed to allocate page for rx qps\n");
3311 		goto out;
3312 	}
3313 	priv->allocated = 1;
3314 
3315 	/*
3316 	 * Initialize netdev entry points
3317 	 */
3318 	if (mlx4_is_master(priv->mdev->dev))
3319 		dev->netdev_ops = &mlx4_netdev_ops_master;
3320 	else
3321 		dev->netdev_ops = &mlx4_netdev_ops;
3322 	dev->xdp_metadata_ops = &mlx4_xdp_metadata_ops;
3323 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3324 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3325 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3326 
3327 	dev->stat_ops = &mlx4_stat_ops;
3328 	dev->ethtool_ops = &mlx4_en_ethtool_ops;
3329 
3330 	/*
3331 	 * Set driver features
3332 	 */
3333 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3334 	if (mdev->LSO_support)
3335 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3336 
3337 	if (mdev->dev->caps.tunnel_offload_mode ==
3338 	    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3339 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3340 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3341 				    NETIF_F_GSO_PARTIAL;
3342 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3343 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3344 				    NETIF_F_GSO_PARTIAL;
3345 		dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3346 		dev->hw_enc_features = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3347 				       NETIF_F_RXCSUM |
3348 				       NETIF_F_TSO | NETIF_F_TSO6 |
3349 				       NETIF_F_GSO_UDP_TUNNEL |
3350 				       NETIF_F_GSO_UDP_TUNNEL_CSUM |
3351 				       NETIF_F_GSO_PARTIAL;
3352 
3353 		dev->udp_tunnel_nic_info = &mlx4_udp_tunnels;
3354 	}
3355 
3356 	dev->vlan_features = dev->hw_features;
3357 
3358 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3359 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3360 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3361 			NETIF_F_HW_VLAN_CTAG_FILTER;
3362 	dev->hw_features |= NETIF_F_LOOPBACK |
3363 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3364 
3365 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3366 		dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3367 			NETIF_F_HW_VLAN_STAG_FILTER;
3368 		dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3369 	}
3370 
3371 	if (mlx4_is_slave(mdev->dev)) {
3372 		bool vlan_offload_disabled;
3373 		int phv;
3374 
3375 		err = get_phv_bit(mdev->dev, port, &phv);
3376 		if (!err && phv) {
3377 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3378 			priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3379 		}
3380 		err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3381 							&vlan_offload_disabled);
3382 		if (!err && vlan_offload_disabled) {
3383 			dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3384 					      NETIF_F_HW_VLAN_CTAG_RX |
3385 					      NETIF_F_HW_VLAN_STAG_TX |
3386 					      NETIF_F_HW_VLAN_STAG_RX);
3387 			dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3388 					   NETIF_F_HW_VLAN_CTAG_RX |
3389 					   NETIF_F_HW_VLAN_STAG_TX |
3390 					   NETIF_F_HW_VLAN_STAG_RX);
3391 		}
3392 	} else {
3393 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3394 		    !(mdev->dev->caps.flags2 &
3395 		      MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3396 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3397 	}
3398 
3399 	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3400 		dev->hw_features |= NETIF_F_RXFCS;
3401 
3402 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3403 		dev->hw_features |= NETIF_F_RXALL;
3404 
3405 	if (mdev->dev->caps.steering_mode ==
3406 	    MLX4_STEERING_MODE_DEVICE_MANAGED &&
3407 	    mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3408 		dev->hw_features |= NETIF_F_NTUPLE;
3409 
3410 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3411 		dev->priv_flags |= IFF_UNICAST_FLT;
3412 
3413 	/* Setting a default hash function value */
3414 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3415 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3416 	} else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3417 		priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3418 	} else {
3419 		en_warn(priv,
3420 			"No RSS hash capabilities exposed, using Toeplitz\n");
3421 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3422 	}
3423 
3424 	dev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
3425 
3426 	/* MTU range: 68 - hw-specific max */
3427 	dev->min_mtu = ETH_MIN_MTU;
3428 	dev->max_mtu = priv->max_mtu;
3429 
3430 	/* supports LSOv2 packets. */
3431 	netif_set_tso_max_size(dev, GSO_MAX_SIZE);
3432 
3433 	mdev->pndev[port] = dev;
3434 	mdev->upper[port] = NULL;
3435 
3436 	netif_carrier_off(dev);
3437 	mlx4_en_set_default_moderation(priv);
3438 
3439 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3440 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3441 
3442 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3443 
3444 	/* Configure port */
3445 	mlx4_en_calc_rx_buf(dev);
3446 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3447 				    priv->rx_skb_size + ETH_FCS_LEN,
3448 				    prof->tx_pause, prof->tx_ppp,
3449 				    prof->rx_pause, prof->rx_ppp);
3450 	if (err) {
3451 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3452 		       priv->port, err);
3453 		goto out;
3454 	}
3455 
3456 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3457 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3458 		if (err) {
3459 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3460 			       err);
3461 			goto out;
3462 		}
3463 	}
3464 
3465 	/* Init port */
3466 	en_warn(priv, "Initializing port\n");
3467 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
3468 	if (err) {
3469 		en_err(priv, "Failed Initializing port\n");
3470 		goto out;
3471 	}
3472 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3473 
3474 	/* Initialize time stamp mechanism */
3475 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3476 		mlx4_en_init_timestamp(mdev);
3477 
3478 	queue_delayed_work(mdev->workqueue, &priv->service_task,
3479 			   SERVICE_TASK_DELAY);
3480 
3481 	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3482 				 mdev->profile.prof[priv->port].rx_ppp,
3483 				 mdev->profile.prof[priv->port].rx_pause,
3484 				 mdev->profile.prof[priv->port].tx_ppp,
3485 				 mdev->profile.prof[priv->port].tx_pause);
3486 
3487 	SET_NETDEV_DEVLINK_PORT(dev,
3488 				mlx4_get_devlink_port(mdev->dev, priv->port));
3489 	err = register_netdev(dev);
3490 	if (err) {
3491 		en_err(priv, "Netdev registration failed for port %d\n", port);
3492 		goto out;
3493 	}
3494 
3495 	priv->registered = 1;
3496 
3497 	return 0;
3498 
3499 out:
3500 	mlx4_en_destroy_netdev(dev);
3501 	return err;
3502 }
3503 
3504 int mlx4_en_reset_config(struct net_device *dev,
3505 			 struct kernel_hwtstamp_config *ts_config,
3506 			 netdev_features_t features)
3507 {
3508 	struct mlx4_en_priv *priv = netdev_priv(dev);
3509 	struct mlx4_en_dev *mdev = priv->mdev;
3510 	struct mlx4_en_port_profile new_prof;
3511 	struct mlx4_en_priv *tmp;
3512 	int port_up = 0;
3513 	int err = 0;
3514 
3515 	if (priv->hwtstamp_config.tx_type == ts_config->tx_type &&
3516 	    priv->hwtstamp_config.rx_filter == ts_config->rx_filter &&
3517 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3518 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3519 		return 0; /* Nothing to change */
3520 
3521 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3522 	    (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3523 	    (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3524 		en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3525 		return -EINVAL;
3526 	}
3527 
3528 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3529 	if (!tmp)
3530 		return -ENOMEM;
3531 
3532 	mutex_lock(&mdev->state_lock);
3533 
3534 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3535 	memcpy(&new_prof.hwtstamp_config, ts_config, sizeof(*ts_config));
3536 
3537 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3538 	if (err)
3539 		goto out;
3540 
3541 	if (priv->port_up) {
3542 		port_up = 1;
3543 		mlx4_en_stop_port(dev, 1);
3544 	}
3545 
3546 	mlx4_en_safe_replace_resources(priv, tmp);
3547 
3548 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3549 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3550 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3551 		else
3552 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3553 	} else if (ts_config->rx_filter == HWTSTAMP_FILTER_NONE) {
3554 		/* RX time-stamping is OFF, update the RX vlan offload
3555 		 * to the latest wanted state
3556 		 */
3557 		if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3558 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3559 		else
3560 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3561 	}
3562 
3563 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3564 		if (features & NETIF_F_RXFCS)
3565 			dev->features |= NETIF_F_RXFCS;
3566 		else
3567 			dev->features &= ~NETIF_F_RXFCS;
3568 	}
3569 
3570 	/* RX vlan offload and RX time-stamping can't co-exist !
3571 	 * Regardless of the caller's choice,
3572 	 * Turn Off RX vlan offload in case of time-stamping is ON
3573 	 */
3574 	if (ts_config->rx_filter != HWTSTAMP_FILTER_NONE) {
3575 		if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3576 			en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3577 		dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3578 	}
3579 
3580 	if (port_up) {
3581 		err = mlx4_en_start_port(dev);
3582 		if (err)
3583 			en_err(priv, "Failed starting port\n");
3584 	}
3585 
3586 	if (!err)
3587 		err = mlx4_en_moderation_update(priv);
3588 out:
3589 	mutex_unlock(&mdev->state_lock);
3590 	kfree(tmp);
3591 	if (!err)
3592 		netdev_features_change(dev);
3593 	return err;
3594 }
3595