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/tcp.h>
37 #include <linux/if_vlan.h>
38 #include <linux/delay.h>
39 #include <linux/slab.h>
40 #include <linux/hash.h>
41 #include <net/ip.h>
42 #include <net/vxlan.h>
43 #include <net/devlink.h>
44 
45 #include <linux/mlx4/driver.h>
46 #include <linux/mlx4/device.h>
47 #include <linux/mlx4/cmd.h>
48 #include <linux/mlx4/cq.h>
49 
50 #include "mlx4_en.h"
51 #include "en_port.h"
52 
53 #define MLX4_EN_MAX_XDP_MTU ((int)(PAGE_SIZE - ETH_HLEN - (2 * VLAN_HLEN) - \
54 				   XDP_PACKET_HEADROOM))
55 
56 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
57 {
58 	struct mlx4_en_priv *priv = netdev_priv(dev);
59 	int i;
60 	unsigned int offset = 0;
61 
62 	if (up && up != MLX4_EN_NUM_UP_HIGH)
63 		return -EINVAL;
64 
65 	netdev_set_num_tc(dev, up);
66 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
67 	/* Partition Tx queues evenly amongst UP's */
68 	for (i = 0; i < up; i++) {
69 		netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
70 		offset += priv->num_tx_rings_p_up;
71 	}
72 
73 #ifdef CONFIG_MLX4_EN_DCB
74 	if (!mlx4_is_slave(priv->mdev->dev)) {
75 		if (up) {
76 			if (priv->dcbx_cap)
77 				priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
78 		} else {
79 			priv->flags &= ~MLX4_EN_FLAG_DCB_ENABLED;
80 			priv->cee_config.pfc_state = false;
81 		}
82 	}
83 #endif /* CONFIG_MLX4_EN_DCB */
84 
85 	return 0;
86 }
87 
88 int mlx4_en_alloc_tx_queue_per_tc(struct net_device *dev, u8 tc)
89 {
90 	struct mlx4_en_priv *priv = netdev_priv(dev);
91 	struct mlx4_en_dev *mdev = priv->mdev;
92 	struct mlx4_en_port_profile new_prof;
93 	struct mlx4_en_priv *tmp;
94 	int port_up = 0;
95 	int err = 0;
96 
97 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
98 	if (!tmp)
99 		return -ENOMEM;
100 
101 	mutex_lock(&mdev->state_lock);
102 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
103 	new_prof.num_up = (tc == 0) ? MLX4_EN_NUM_UP_LOW :
104 				      MLX4_EN_NUM_UP_HIGH;
105 	new_prof.tx_ring_num[TX] = new_prof.num_tx_rings_p_up *
106 				   new_prof.num_up;
107 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
108 	if (err)
109 		goto out;
110 
111 	if (priv->port_up) {
112 		port_up = 1;
113 		mlx4_en_stop_port(dev, 1);
114 	}
115 
116 	mlx4_en_safe_replace_resources(priv, tmp);
117 	if (port_up) {
118 		err = mlx4_en_start_port(dev);
119 		if (err) {
120 			en_err(priv, "Failed starting port for setup TC\n");
121 			goto out;
122 		}
123 	}
124 
125 	err = mlx4_en_setup_tc(dev, tc);
126 out:
127 	mutex_unlock(&mdev->state_lock);
128 	kfree(tmp);
129 	return err;
130 }
131 
132 static int __mlx4_en_setup_tc(struct net_device *dev, enum tc_setup_type type,
133 			      void *type_data)
134 {
135 	struct tc_mqprio_qopt *mqprio = type_data;
136 
137 	if (type != TC_SETUP_QDISC_MQPRIO)
138 		return -EOPNOTSUPP;
139 
140 	if (mqprio->num_tc && mqprio->num_tc != MLX4_EN_NUM_UP_HIGH)
141 		return -EINVAL;
142 
143 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
144 
145 	return mlx4_en_alloc_tx_queue_per_tc(dev, mqprio->num_tc);
146 }
147 
148 #ifdef CONFIG_RFS_ACCEL
149 
150 struct mlx4_en_filter {
151 	struct list_head next;
152 	struct work_struct work;
153 
154 	u8     ip_proto;
155 	__be32 src_ip;
156 	__be32 dst_ip;
157 	__be16 src_port;
158 	__be16 dst_port;
159 
160 	int rxq_index;
161 	struct mlx4_en_priv *priv;
162 	u32 flow_id;			/* RFS infrastructure id */
163 	int id;				/* mlx4_en driver id */
164 	u64 reg_id;			/* Flow steering API id */
165 	u8 activated;			/* Used to prevent expiry before filter
166 					 * is attached
167 					 */
168 	struct hlist_node filter_chain;
169 };
170 
171 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
172 
173 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
174 {
175 	switch (ip_proto) {
176 	case IPPROTO_UDP:
177 		return MLX4_NET_TRANS_RULE_ID_UDP;
178 	case IPPROTO_TCP:
179 		return MLX4_NET_TRANS_RULE_ID_TCP;
180 	default:
181 		return MLX4_NET_TRANS_RULE_NUM;
182 	}
183 };
184 
185 /* Must not acquire state_lock, as its corresponding work_sync
186  * is done under it.
187  */
188 static void mlx4_en_filter_work(struct work_struct *work)
189 {
190 	struct mlx4_en_filter *filter = container_of(work,
191 						     struct mlx4_en_filter,
192 						     work);
193 	struct mlx4_en_priv *priv = filter->priv;
194 	struct mlx4_spec_list spec_tcp_udp = {
195 		.id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
196 		{
197 			.tcp_udp = {
198 				.dst_port = filter->dst_port,
199 				.dst_port_msk = (__force __be16)-1,
200 				.src_port = filter->src_port,
201 				.src_port_msk = (__force __be16)-1,
202 			},
203 		},
204 	};
205 	struct mlx4_spec_list spec_ip = {
206 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
207 		{
208 			.ipv4 = {
209 				.dst_ip = filter->dst_ip,
210 				.dst_ip_msk = (__force __be32)-1,
211 				.src_ip = filter->src_ip,
212 				.src_ip_msk = (__force __be32)-1,
213 			},
214 		},
215 	};
216 	struct mlx4_spec_list spec_eth = {
217 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
218 	};
219 	struct mlx4_net_trans_rule rule = {
220 		.list = LIST_HEAD_INIT(rule.list),
221 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
222 		.exclusive = 1,
223 		.allow_loopback = 1,
224 		.promisc_mode = MLX4_FS_REGULAR,
225 		.port = priv->port,
226 		.priority = MLX4_DOMAIN_RFS,
227 	};
228 	int rc;
229 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
230 
231 	if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
232 		en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
233 			filter->ip_proto);
234 		goto ignore;
235 	}
236 	list_add_tail(&spec_eth.list, &rule.list);
237 	list_add_tail(&spec_ip.list, &rule.list);
238 	list_add_tail(&spec_tcp_udp.list, &rule.list);
239 
240 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
241 	memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
242 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
243 
244 	filter->activated = 0;
245 
246 	if (filter->reg_id) {
247 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
248 		if (rc && rc != -ENOENT)
249 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
250 	}
251 
252 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
253 	if (rc)
254 		en_err(priv, "Error attaching flow. err = %d\n", rc);
255 
256 ignore:
257 	mlx4_en_filter_rfs_expire(priv);
258 
259 	filter->activated = 1;
260 }
261 
262 static inline struct hlist_head *
263 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
264 		   __be16 src_port, __be16 dst_port)
265 {
266 	unsigned long l;
267 	int bucket_idx;
268 
269 	l = (__force unsigned long)src_port |
270 	    ((__force unsigned long)dst_port << 2);
271 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
272 
273 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
274 
275 	return &priv->filter_hash[bucket_idx];
276 }
277 
278 static struct mlx4_en_filter *
279 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
280 		     __be32 dst_ip, u8 ip_proto, __be16 src_port,
281 		     __be16 dst_port, u32 flow_id)
282 {
283 	struct mlx4_en_filter *filter = NULL;
284 
285 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
286 	if (!filter)
287 		return NULL;
288 
289 	filter->priv = priv;
290 	filter->rxq_index = rxq_index;
291 	INIT_WORK(&filter->work, mlx4_en_filter_work);
292 
293 	filter->src_ip = src_ip;
294 	filter->dst_ip = dst_ip;
295 	filter->ip_proto = ip_proto;
296 	filter->src_port = src_port;
297 	filter->dst_port = dst_port;
298 
299 	filter->flow_id = flow_id;
300 
301 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
302 
303 	list_add_tail(&filter->next, &priv->filters);
304 	hlist_add_head(&filter->filter_chain,
305 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
306 					  dst_port));
307 
308 	return filter;
309 }
310 
311 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
312 {
313 	struct mlx4_en_priv *priv = filter->priv;
314 	int rc;
315 
316 	list_del(&filter->next);
317 
318 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
319 	if (rc && rc != -ENOENT)
320 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
321 
322 	kfree(filter);
323 }
324 
325 static inline struct mlx4_en_filter *
326 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
327 		    u8 ip_proto, __be16 src_port, __be16 dst_port)
328 {
329 	struct mlx4_en_filter *filter;
330 	struct mlx4_en_filter *ret = NULL;
331 
332 	hlist_for_each_entry(filter,
333 			     filter_hash_bucket(priv, src_ip, dst_ip,
334 						src_port, dst_port),
335 			     filter_chain) {
336 		if (filter->src_ip == src_ip &&
337 		    filter->dst_ip == dst_ip &&
338 		    filter->ip_proto == ip_proto &&
339 		    filter->src_port == src_port &&
340 		    filter->dst_port == dst_port) {
341 			ret = filter;
342 			break;
343 		}
344 	}
345 
346 	return ret;
347 }
348 
349 static int
350 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
351 		   u16 rxq_index, u32 flow_id)
352 {
353 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
354 	struct mlx4_en_filter *filter;
355 	const struct iphdr *ip;
356 	const __be16 *ports;
357 	u8 ip_proto;
358 	__be32 src_ip;
359 	__be32 dst_ip;
360 	__be16 src_port;
361 	__be16 dst_port;
362 	int nhoff = skb_network_offset(skb);
363 	int ret = 0;
364 
365 	if (skb->protocol != htons(ETH_P_IP))
366 		return -EPROTONOSUPPORT;
367 
368 	ip = (const struct iphdr *)(skb->data + nhoff);
369 	if (ip_is_fragment(ip))
370 		return -EPROTONOSUPPORT;
371 
372 	if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
373 		return -EPROTONOSUPPORT;
374 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
375 
376 	ip_proto = ip->protocol;
377 	src_ip = ip->saddr;
378 	dst_ip = ip->daddr;
379 	src_port = ports[0];
380 	dst_port = ports[1];
381 
382 	spin_lock_bh(&priv->filters_lock);
383 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
384 				     src_port, dst_port);
385 	if (filter) {
386 		if (filter->rxq_index == rxq_index)
387 			goto out;
388 
389 		filter->rxq_index = rxq_index;
390 	} else {
391 		filter = mlx4_en_filter_alloc(priv, rxq_index,
392 					      src_ip, dst_ip, ip_proto,
393 					      src_port, dst_port, flow_id);
394 		if (!filter) {
395 			ret = -ENOMEM;
396 			goto err;
397 		}
398 	}
399 
400 	queue_work(priv->mdev->workqueue, &filter->work);
401 
402 out:
403 	ret = filter->id;
404 err:
405 	spin_unlock_bh(&priv->filters_lock);
406 
407 	return ret;
408 }
409 
410 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
411 {
412 	struct mlx4_en_filter *filter, *tmp;
413 	LIST_HEAD(del_list);
414 
415 	spin_lock_bh(&priv->filters_lock);
416 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
417 		list_move(&filter->next, &del_list);
418 		hlist_del(&filter->filter_chain);
419 	}
420 	spin_unlock_bh(&priv->filters_lock);
421 
422 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
423 		cancel_work_sync(&filter->work);
424 		mlx4_en_filter_free(filter);
425 	}
426 }
427 
428 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
429 {
430 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
431 	LIST_HEAD(del_list);
432 	int i = 0;
433 
434 	spin_lock_bh(&priv->filters_lock);
435 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
436 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
437 			break;
438 
439 		if (filter->activated &&
440 		    !work_pending(&filter->work) &&
441 		    rps_may_expire_flow(priv->dev,
442 					filter->rxq_index, filter->flow_id,
443 					filter->id)) {
444 			list_move(&filter->next, &del_list);
445 			hlist_del(&filter->filter_chain);
446 		} else
447 			last_filter = filter;
448 
449 		i++;
450 	}
451 
452 	if (last_filter && (&last_filter->next != priv->filters.next))
453 		list_move(&priv->filters, &last_filter->next);
454 
455 	spin_unlock_bh(&priv->filters_lock);
456 
457 	list_for_each_entry_safe(filter, tmp, &del_list, next)
458 		mlx4_en_filter_free(filter);
459 }
460 #endif
461 
462 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
463 				   __be16 proto, u16 vid)
464 {
465 	struct mlx4_en_priv *priv = netdev_priv(dev);
466 	struct mlx4_en_dev *mdev = priv->mdev;
467 	int err;
468 	int idx;
469 
470 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
471 
472 	set_bit(vid, priv->active_vlans);
473 
474 	/* Add VID to port VLAN filter */
475 	mutex_lock(&mdev->state_lock);
476 	if (mdev->device_up && priv->port_up) {
477 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
478 		if (err) {
479 			en_err(priv, "Failed configuring VLAN filter\n");
480 			goto out;
481 		}
482 	}
483 	err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
484 	if (err)
485 		en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
486 
487 out:
488 	mutex_unlock(&mdev->state_lock);
489 	return err;
490 }
491 
492 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
493 				    __be16 proto, u16 vid)
494 {
495 	struct mlx4_en_priv *priv = netdev_priv(dev);
496 	struct mlx4_en_dev *mdev = priv->mdev;
497 	int err = 0;
498 
499 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
500 
501 	clear_bit(vid, priv->active_vlans);
502 
503 	/* Remove VID from port VLAN filter */
504 	mutex_lock(&mdev->state_lock);
505 	mlx4_unregister_vlan(mdev->dev, priv->port, vid);
506 
507 	if (mdev->device_up && priv->port_up) {
508 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
509 		if (err)
510 			en_err(priv, "Failed configuring VLAN filter\n");
511 	}
512 	mutex_unlock(&mdev->state_lock);
513 
514 	return err;
515 }
516 
517 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
518 {
519 	int i;
520 	for (i = ETH_ALEN - 1; i >= 0; --i) {
521 		dst_mac[i] = src_mac & 0xff;
522 		src_mac >>= 8;
523 	}
524 	memset(&dst_mac[ETH_ALEN], 0, 2);
525 }
526 
527 
528 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
529 				    int qpn, u64 *reg_id)
530 {
531 	int err;
532 
533 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
534 	    priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
535 		return 0; /* do nothing */
536 
537 	err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
538 				    MLX4_DOMAIN_NIC, reg_id);
539 	if (err) {
540 		en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
541 		return err;
542 	}
543 	en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
544 	return 0;
545 }
546 
547 
548 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
549 				unsigned char *mac, int *qpn, u64 *reg_id)
550 {
551 	struct mlx4_en_dev *mdev = priv->mdev;
552 	struct mlx4_dev *dev = mdev->dev;
553 	int err;
554 
555 	switch (dev->caps.steering_mode) {
556 	case MLX4_STEERING_MODE_B0: {
557 		struct mlx4_qp qp;
558 		u8 gid[16] = {0};
559 
560 		qp.qpn = *qpn;
561 		memcpy(&gid[10], mac, ETH_ALEN);
562 		gid[5] = priv->port;
563 
564 		err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
565 		break;
566 	}
567 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
568 		struct mlx4_spec_list spec_eth = { {NULL} };
569 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
570 
571 		struct mlx4_net_trans_rule rule = {
572 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
573 			.exclusive = 0,
574 			.allow_loopback = 1,
575 			.promisc_mode = MLX4_FS_REGULAR,
576 			.priority = MLX4_DOMAIN_NIC,
577 		};
578 
579 		rule.port = priv->port;
580 		rule.qpn = *qpn;
581 		INIT_LIST_HEAD(&rule.list);
582 
583 		spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
584 		memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
585 		memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
586 		list_add_tail(&spec_eth.list, &rule.list);
587 
588 		err = mlx4_flow_attach(dev, &rule, reg_id);
589 		break;
590 	}
591 	default:
592 		return -EINVAL;
593 	}
594 	if (err)
595 		en_warn(priv, "Failed Attaching Unicast\n");
596 
597 	return err;
598 }
599 
600 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
601 				     unsigned char *mac, int qpn, u64 reg_id)
602 {
603 	struct mlx4_en_dev *mdev = priv->mdev;
604 	struct mlx4_dev *dev = mdev->dev;
605 
606 	switch (dev->caps.steering_mode) {
607 	case MLX4_STEERING_MODE_B0: {
608 		struct mlx4_qp qp;
609 		u8 gid[16] = {0};
610 
611 		qp.qpn = qpn;
612 		memcpy(&gid[10], mac, ETH_ALEN);
613 		gid[5] = priv->port;
614 
615 		mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
616 		break;
617 	}
618 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
619 		mlx4_flow_detach(dev, reg_id);
620 		break;
621 	}
622 	default:
623 		en_err(priv, "Invalid steering mode.\n");
624 	}
625 }
626 
627 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
628 {
629 	struct mlx4_en_dev *mdev = priv->mdev;
630 	struct mlx4_dev *dev = mdev->dev;
631 	int index = 0;
632 	int err = 0;
633 	int *qpn = &priv->base_qpn;
634 	u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
635 
636 	en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
637 	       priv->dev->dev_addr);
638 	index = mlx4_register_mac(dev, priv->port, mac);
639 	if (index < 0) {
640 		err = index;
641 		en_err(priv, "Failed adding MAC: %pM\n",
642 		       priv->dev->dev_addr);
643 		return err;
644 	}
645 
646 	en_info(priv, "Steering Mode %d\n", dev->caps.steering_mode);
647 
648 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
649 		int base_qpn = mlx4_get_base_qpn(dev, priv->port);
650 		*qpn = base_qpn + index;
651 		return 0;
652 	}
653 
654 	err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP,
655 				    MLX4_RES_USAGE_DRIVER);
656 	en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
657 	if (err) {
658 		en_err(priv, "Failed to reserve qp for mac registration\n");
659 		mlx4_unregister_mac(dev, priv->port, mac);
660 		return err;
661 	}
662 
663 	return 0;
664 }
665 
666 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
667 {
668 	struct mlx4_en_dev *mdev = priv->mdev;
669 	struct mlx4_dev *dev = mdev->dev;
670 	int qpn = priv->base_qpn;
671 
672 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
673 		u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
674 		en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
675 		       priv->dev->dev_addr);
676 		mlx4_unregister_mac(dev, priv->port, mac);
677 	} else {
678 		en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
679 		       priv->port, qpn);
680 		mlx4_qp_release_range(dev, qpn, 1);
681 		priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
682 	}
683 }
684 
685 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
686 			       unsigned char *new_mac, unsigned char *prev_mac)
687 {
688 	struct mlx4_en_dev *mdev = priv->mdev;
689 	struct mlx4_dev *dev = mdev->dev;
690 	int err = 0;
691 	u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
692 
693 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
694 		struct hlist_head *bucket;
695 		unsigned int mac_hash;
696 		struct mlx4_mac_entry *entry;
697 		struct hlist_node *tmp;
698 		u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
699 
700 		bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
701 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
702 			if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
703 				mlx4_en_uc_steer_release(priv, entry->mac,
704 							 qpn, entry->reg_id);
705 				mlx4_unregister_mac(dev, priv->port,
706 						    prev_mac_u64);
707 				hlist_del_rcu(&entry->hlist);
708 				synchronize_rcu();
709 				memcpy(entry->mac, new_mac, ETH_ALEN);
710 				entry->reg_id = 0;
711 				mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
712 				hlist_add_head_rcu(&entry->hlist,
713 						   &priv->mac_hash[mac_hash]);
714 				mlx4_register_mac(dev, priv->port, new_mac_u64);
715 				err = mlx4_en_uc_steer_add(priv, new_mac,
716 							   &qpn,
717 							   &entry->reg_id);
718 				if (err)
719 					return err;
720 				if (priv->tunnel_reg_id) {
721 					mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
722 					priv->tunnel_reg_id = 0;
723 				}
724 				err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
725 							       &priv->tunnel_reg_id);
726 				return err;
727 			}
728 		}
729 		return -EINVAL;
730 	}
731 
732 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
733 }
734 
735 static void mlx4_en_update_user_mac(struct mlx4_en_priv *priv,
736 				    unsigned char new_mac[ETH_ALEN + 2])
737 {
738 	struct mlx4_en_dev *mdev = priv->mdev;
739 	int err;
740 
741 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_USER_MAC_EN))
742 		return;
743 
744 	err = mlx4_SET_PORT_user_mac(mdev->dev, priv->port, new_mac);
745 	if (err)
746 		en_err(priv, "Failed to pass user MAC(%pM) to Firmware for port %d, with error %d\n",
747 		       new_mac, priv->port, err);
748 }
749 
750 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
751 			      unsigned char new_mac[ETH_ALEN + 2])
752 {
753 	int err = 0;
754 
755 	if (priv->port_up) {
756 		/* Remove old MAC and insert the new one */
757 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
758 					  new_mac, priv->current_mac);
759 		if (err)
760 			en_err(priv, "Failed changing HW MAC address\n");
761 	} else
762 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
763 
764 	if (!err)
765 		memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
766 
767 	return err;
768 }
769 
770 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
771 {
772 	struct mlx4_en_priv *priv = netdev_priv(dev);
773 	struct mlx4_en_dev *mdev = priv->mdev;
774 	struct sockaddr *saddr = addr;
775 	unsigned char new_mac[ETH_ALEN + 2];
776 	int err;
777 
778 	if (!is_valid_ether_addr(saddr->sa_data))
779 		return -EADDRNOTAVAIL;
780 
781 	mutex_lock(&mdev->state_lock);
782 	memcpy(new_mac, saddr->sa_data, ETH_ALEN);
783 	err = mlx4_en_do_set_mac(priv, new_mac);
784 	if (err)
785 		goto out;
786 
787 	memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
788 	mlx4_en_update_user_mac(priv, new_mac);
789 out:
790 	mutex_unlock(&mdev->state_lock);
791 
792 	return err;
793 }
794 
795 static void mlx4_en_clear_list(struct net_device *dev)
796 {
797 	struct mlx4_en_priv *priv = netdev_priv(dev);
798 	struct mlx4_en_mc_list *tmp, *mc_to_del;
799 
800 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
801 		list_del(&mc_to_del->list);
802 		kfree(mc_to_del);
803 	}
804 }
805 
806 static void mlx4_en_cache_mclist(struct net_device *dev)
807 {
808 	struct mlx4_en_priv *priv = netdev_priv(dev);
809 	struct netdev_hw_addr *ha;
810 	struct mlx4_en_mc_list *tmp;
811 
812 	mlx4_en_clear_list(dev);
813 	netdev_for_each_mc_addr(ha, dev) {
814 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
815 		if (!tmp) {
816 			mlx4_en_clear_list(dev);
817 			return;
818 		}
819 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
820 		list_add_tail(&tmp->list, &priv->mc_list);
821 	}
822 }
823 
824 static void update_mclist_flags(struct mlx4_en_priv *priv,
825 				struct list_head *dst,
826 				struct list_head *src)
827 {
828 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
829 	bool found;
830 
831 	/* Find all the entries that should be removed from dst,
832 	 * These are the entries that are not found in src
833 	 */
834 	list_for_each_entry(dst_tmp, dst, list) {
835 		found = false;
836 		list_for_each_entry(src_tmp, src, list) {
837 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
838 				found = true;
839 				break;
840 			}
841 		}
842 		if (!found)
843 			dst_tmp->action = MCLIST_REM;
844 	}
845 
846 	/* Add entries that exist in src but not in dst
847 	 * mark them as need to add
848 	 */
849 	list_for_each_entry(src_tmp, src, list) {
850 		found = false;
851 		list_for_each_entry(dst_tmp, dst, list) {
852 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
853 				dst_tmp->action = MCLIST_NONE;
854 				found = true;
855 				break;
856 			}
857 		}
858 		if (!found) {
859 			new_mc = kmemdup(src_tmp,
860 					 sizeof(struct mlx4_en_mc_list),
861 					 GFP_KERNEL);
862 			if (!new_mc)
863 				return;
864 
865 			new_mc->action = MCLIST_ADD;
866 			list_add_tail(&new_mc->list, dst);
867 		}
868 	}
869 }
870 
871 static void mlx4_en_set_rx_mode(struct net_device *dev)
872 {
873 	struct mlx4_en_priv *priv = netdev_priv(dev);
874 
875 	if (!priv->port_up)
876 		return;
877 
878 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
879 }
880 
881 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
882 				     struct mlx4_en_dev *mdev)
883 {
884 	int err = 0;
885 
886 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
887 		if (netif_msg_rx_status(priv))
888 			en_warn(priv, "Entering promiscuous mode\n");
889 		priv->flags |= MLX4_EN_FLAG_PROMISC;
890 
891 		/* Enable promiscouos mode */
892 		switch (mdev->dev->caps.steering_mode) {
893 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
894 			err = mlx4_flow_steer_promisc_add(mdev->dev,
895 							  priv->port,
896 							  priv->base_qpn,
897 							  MLX4_FS_ALL_DEFAULT);
898 			if (err)
899 				en_err(priv, "Failed enabling promiscuous mode\n");
900 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
901 			break;
902 
903 		case MLX4_STEERING_MODE_B0:
904 			err = mlx4_unicast_promisc_add(mdev->dev,
905 						       priv->base_qpn,
906 						       priv->port);
907 			if (err)
908 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
909 
910 			/* Add the default qp number as multicast
911 			 * promisc
912 			 */
913 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
914 				err = mlx4_multicast_promisc_add(mdev->dev,
915 								 priv->base_qpn,
916 								 priv->port);
917 				if (err)
918 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
919 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
920 			}
921 			break;
922 
923 		case MLX4_STEERING_MODE_A0:
924 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
925 						     priv->port,
926 						     priv->base_qpn,
927 						     1);
928 			if (err)
929 				en_err(priv, "Failed enabling promiscuous mode\n");
930 			break;
931 		}
932 
933 		/* Disable port multicast filter (unconditionally) */
934 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
935 					  0, MLX4_MCAST_DISABLE);
936 		if (err)
937 			en_err(priv, "Failed disabling multicast filter\n");
938 	}
939 }
940 
941 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
942 				       struct mlx4_en_dev *mdev)
943 {
944 	int err = 0;
945 
946 	if (netif_msg_rx_status(priv))
947 		en_warn(priv, "Leaving promiscuous mode\n");
948 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
949 
950 	/* Disable promiscouos mode */
951 	switch (mdev->dev->caps.steering_mode) {
952 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
953 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
954 						     priv->port,
955 						     MLX4_FS_ALL_DEFAULT);
956 		if (err)
957 			en_err(priv, "Failed disabling promiscuous mode\n");
958 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
959 		break;
960 
961 	case MLX4_STEERING_MODE_B0:
962 		err = mlx4_unicast_promisc_remove(mdev->dev,
963 						  priv->base_qpn,
964 						  priv->port);
965 		if (err)
966 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
967 		/* Disable Multicast promisc */
968 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
969 			err = mlx4_multicast_promisc_remove(mdev->dev,
970 							    priv->base_qpn,
971 							    priv->port);
972 			if (err)
973 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
974 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
975 		}
976 		break;
977 
978 	case MLX4_STEERING_MODE_A0:
979 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
980 					     priv->port,
981 					     priv->base_qpn, 0);
982 		if (err)
983 			en_err(priv, "Failed disabling promiscuous mode\n");
984 		break;
985 	}
986 }
987 
988 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
989 				 struct net_device *dev,
990 				 struct mlx4_en_dev *mdev)
991 {
992 	struct mlx4_en_mc_list *mclist, *tmp;
993 	u64 mcast_addr = 0;
994 	u8 mc_list[16] = {0};
995 	int err = 0;
996 
997 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
998 	if (dev->flags & IFF_ALLMULTI) {
999 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1000 					  0, MLX4_MCAST_DISABLE);
1001 		if (err)
1002 			en_err(priv, "Failed disabling multicast filter\n");
1003 
1004 		/* Add the default qp number as multicast promisc */
1005 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1006 			switch (mdev->dev->caps.steering_mode) {
1007 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1008 				err = mlx4_flow_steer_promisc_add(mdev->dev,
1009 								  priv->port,
1010 								  priv->base_qpn,
1011 								  MLX4_FS_MC_DEFAULT);
1012 				break;
1013 
1014 			case MLX4_STEERING_MODE_B0:
1015 				err = mlx4_multicast_promisc_add(mdev->dev,
1016 								 priv->base_qpn,
1017 								 priv->port);
1018 				break;
1019 
1020 			case MLX4_STEERING_MODE_A0:
1021 				break;
1022 			}
1023 			if (err)
1024 				en_err(priv, "Failed entering multicast promisc mode\n");
1025 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1026 		}
1027 	} else {
1028 		/* Disable Multicast promisc */
1029 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1030 			switch (mdev->dev->caps.steering_mode) {
1031 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1032 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
1033 								     priv->port,
1034 								     MLX4_FS_MC_DEFAULT);
1035 				break;
1036 
1037 			case MLX4_STEERING_MODE_B0:
1038 				err = mlx4_multicast_promisc_remove(mdev->dev,
1039 								    priv->base_qpn,
1040 								    priv->port);
1041 				break;
1042 
1043 			case MLX4_STEERING_MODE_A0:
1044 				break;
1045 			}
1046 			if (err)
1047 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
1048 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1049 		}
1050 
1051 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1052 					  0, MLX4_MCAST_DISABLE);
1053 		if (err)
1054 			en_err(priv, "Failed disabling multicast filter\n");
1055 
1056 		/* Flush mcast filter and init it with broadcast address */
1057 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1058 				    1, MLX4_MCAST_CONFIG);
1059 
1060 		/* Update multicast list - we cache all addresses so they won't
1061 		 * change while HW is updated holding the command semaphor */
1062 		netif_addr_lock_bh(dev);
1063 		mlx4_en_cache_mclist(dev);
1064 		netif_addr_unlock_bh(dev);
1065 		list_for_each_entry(mclist, &priv->mc_list, list) {
1066 			mcast_addr = mlx4_mac_to_u64(mclist->addr);
1067 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1068 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
1069 		}
1070 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1071 					  0, MLX4_MCAST_ENABLE);
1072 		if (err)
1073 			en_err(priv, "Failed enabling multicast filter\n");
1074 
1075 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1076 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1077 			if (mclist->action == MCLIST_REM) {
1078 				/* detach this address and delete from list */
1079 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1080 				mc_list[5] = priv->port;
1081 				err = mlx4_multicast_detach(mdev->dev,
1082 							    priv->rss_map.indir_qp,
1083 							    mc_list,
1084 							    MLX4_PROT_ETH,
1085 							    mclist->reg_id);
1086 				if (err)
1087 					en_err(priv, "Fail to detach multicast address\n");
1088 
1089 				if (mclist->tunnel_reg_id) {
1090 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1091 					if (err)
1092 						en_err(priv, "Failed to detach multicast address\n");
1093 				}
1094 
1095 				/* remove from list */
1096 				list_del(&mclist->list);
1097 				kfree(mclist);
1098 			} else if (mclist->action == MCLIST_ADD) {
1099 				/* attach the address */
1100 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1101 				/* needed for B0 steering support */
1102 				mc_list[5] = priv->port;
1103 				err = mlx4_multicast_attach(mdev->dev,
1104 							    priv->rss_map.indir_qp,
1105 							    mc_list,
1106 							    priv->port, 0,
1107 							    MLX4_PROT_ETH,
1108 							    &mclist->reg_id);
1109 				if (err)
1110 					en_err(priv, "Fail to attach multicast address\n");
1111 
1112 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1113 							       &mclist->tunnel_reg_id);
1114 				if (err)
1115 					en_err(priv, "Failed to attach multicast address\n");
1116 			}
1117 		}
1118 	}
1119 }
1120 
1121 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1122 				 struct net_device *dev,
1123 				 struct mlx4_en_dev *mdev)
1124 {
1125 	struct netdev_hw_addr *ha;
1126 	struct mlx4_mac_entry *entry;
1127 	struct hlist_node *tmp;
1128 	bool found;
1129 	u64 mac;
1130 	int err = 0;
1131 	struct hlist_head *bucket;
1132 	unsigned int i;
1133 	int removed = 0;
1134 	u32 prev_flags;
1135 
1136 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1137 	 * since all modification code is protected by mdev->state_lock
1138 	 */
1139 
1140 	/* find what to remove */
1141 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1142 		bucket = &priv->mac_hash[i];
1143 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1144 			found = false;
1145 			netdev_for_each_uc_addr(ha, dev) {
1146 				if (ether_addr_equal_64bits(entry->mac,
1147 							    ha->addr)) {
1148 					found = true;
1149 					break;
1150 				}
1151 			}
1152 
1153 			/* MAC address of the port is not in uc list */
1154 			if (ether_addr_equal_64bits(entry->mac,
1155 						    priv->current_mac))
1156 				found = true;
1157 
1158 			if (!found) {
1159 				mac = mlx4_mac_to_u64(entry->mac);
1160 				mlx4_en_uc_steer_release(priv, entry->mac,
1161 							 priv->base_qpn,
1162 							 entry->reg_id);
1163 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1164 
1165 				hlist_del_rcu(&entry->hlist);
1166 				kfree_rcu(entry, rcu);
1167 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1168 				       entry->mac, priv->port);
1169 				++removed;
1170 			}
1171 		}
1172 	}
1173 
1174 	/* if we didn't remove anything, there is no use in trying to add
1175 	 * again once we are in a forced promisc mode state
1176 	 */
1177 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1178 		return;
1179 
1180 	prev_flags = priv->flags;
1181 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1182 
1183 	/* find what to add */
1184 	netdev_for_each_uc_addr(ha, dev) {
1185 		found = false;
1186 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1187 		hlist_for_each_entry(entry, bucket, hlist) {
1188 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1189 				found = true;
1190 				break;
1191 			}
1192 		}
1193 
1194 		if (!found) {
1195 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1196 			if (!entry) {
1197 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1198 				       ha->addr, priv->port);
1199 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1200 				break;
1201 			}
1202 			mac = mlx4_mac_to_u64(ha->addr);
1203 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1204 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1205 			if (err < 0) {
1206 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1207 				       ha->addr, priv->port, err);
1208 				kfree(entry);
1209 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1210 				break;
1211 			}
1212 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1213 						   &priv->base_qpn,
1214 						   &entry->reg_id);
1215 			if (err) {
1216 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1217 				       ha->addr, priv->port, err);
1218 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1219 				kfree(entry);
1220 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1221 				break;
1222 			} else {
1223 				unsigned int mac_hash;
1224 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1225 				       ha->addr, priv->port);
1226 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1227 				bucket = &priv->mac_hash[mac_hash];
1228 				hlist_add_head_rcu(&entry->hlist, bucket);
1229 			}
1230 		}
1231 	}
1232 
1233 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1234 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1235 			priv->port);
1236 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1237 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1238 			priv->port);
1239 	}
1240 }
1241 
1242 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1243 {
1244 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1245 						 rx_mode_task);
1246 	struct mlx4_en_dev *mdev = priv->mdev;
1247 	struct net_device *dev = priv->dev;
1248 
1249 	mutex_lock(&mdev->state_lock);
1250 	if (!mdev->device_up) {
1251 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1252 		goto out;
1253 	}
1254 	if (!priv->port_up) {
1255 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1256 		goto out;
1257 	}
1258 
1259 	if (!netif_carrier_ok(dev)) {
1260 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1261 			if (priv->port_state.link_state) {
1262 				priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1263 				netif_carrier_on(dev);
1264 				en_dbg(LINK, priv, "Link Up\n");
1265 			}
1266 		}
1267 	}
1268 
1269 	if (dev->priv_flags & IFF_UNICAST_FLT)
1270 		mlx4_en_do_uc_filter(priv, dev, mdev);
1271 
1272 	/* Promsicuous mode: disable all filters */
1273 	if ((dev->flags & IFF_PROMISC) ||
1274 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1275 		mlx4_en_set_promisc_mode(priv, mdev);
1276 		goto out;
1277 	}
1278 
1279 	/* Not in promiscuous mode */
1280 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1281 		mlx4_en_clear_promisc_mode(priv, mdev);
1282 
1283 	mlx4_en_do_multicast(priv, dev, mdev);
1284 out:
1285 	mutex_unlock(&mdev->state_lock);
1286 }
1287 
1288 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1289 {
1290 	u64 reg_id;
1291 	int err = 0;
1292 	int *qpn = &priv->base_qpn;
1293 	struct mlx4_mac_entry *entry;
1294 
1295 	err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1296 	if (err)
1297 		return err;
1298 
1299 	err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1300 				       &priv->tunnel_reg_id);
1301 	if (err)
1302 		goto tunnel_err;
1303 
1304 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1305 	if (!entry) {
1306 		err = -ENOMEM;
1307 		goto alloc_err;
1308 	}
1309 
1310 	memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1311 	memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1312 	entry->reg_id = reg_id;
1313 	hlist_add_head_rcu(&entry->hlist,
1314 			   &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1315 
1316 	return 0;
1317 
1318 alloc_err:
1319 	if (priv->tunnel_reg_id)
1320 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1321 
1322 tunnel_err:
1323 	mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1324 	return err;
1325 }
1326 
1327 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1328 {
1329 	u64 mac;
1330 	unsigned int i;
1331 	int qpn = priv->base_qpn;
1332 	struct hlist_head *bucket;
1333 	struct hlist_node *tmp;
1334 	struct mlx4_mac_entry *entry;
1335 
1336 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1337 		bucket = &priv->mac_hash[i];
1338 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1339 			mac = mlx4_mac_to_u64(entry->mac);
1340 			en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1341 			       entry->mac);
1342 			mlx4_en_uc_steer_release(priv, entry->mac,
1343 						 qpn, entry->reg_id);
1344 
1345 			mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1346 			hlist_del_rcu(&entry->hlist);
1347 			kfree_rcu(entry, rcu);
1348 		}
1349 	}
1350 
1351 	if (priv->tunnel_reg_id) {
1352 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1353 		priv->tunnel_reg_id = 0;
1354 	}
1355 }
1356 
1357 static void mlx4_en_tx_timeout(struct net_device *dev)
1358 {
1359 	struct mlx4_en_priv *priv = netdev_priv(dev);
1360 	struct mlx4_en_dev *mdev = priv->mdev;
1361 	int i;
1362 
1363 	if (netif_msg_timer(priv))
1364 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1365 
1366 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1367 		struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][i];
1368 
1369 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1370 			continue;
1371 		en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1372 			i, tx_ring->qpn, tx_ring->sp_cqn,
1373 			tx_ring->cons, tx_ring->prod);
1374 	}
1375 
1376 	priv->port_stats.tx_timeout++;
1377 	en_dbg(DRV, priv, "Scheduling watchdog\n");
1378 	queue_work(mdev->workqueue, &priv->watchdog_task);
1379 }
1380 
1381 
1382 static void
1383 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1384 {
1385 	struct mlx4_en_priv *priv = netdev_priv(dev);
1386 
1387 	spin_lock_bh(&priv->stats_lock);
1388 	mlx4_en_fold_software_stats(dev);
1389 	netdev_stats_to_stats64(stats, &dev->stats);
1390 	spin_unlock_bh(&priv->stats_lock);
1391 }
1392 
1393 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1394 {
1395 	struct mlx4_en_cq *cq;
1396 	int i, t;
1397 
1398 	/* If we haven't received a specific coalescing setting
1399 	 * (module param), we set the moderation parameters as follows:
1400 	 * - moder_cnt is set to the number of mtu sized packets to
1401 	 *   satisfy our coalescing target.
1402 	 * - moder_time is set to a fixed value.
1403 	 */
1404 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1405 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1406 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1407 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1408 	en_dbg(INTR, priv, "Default coalescing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1409 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1410 
1411 	/* Setup cq moderation params */
1412 	for (i = 0; i < priv->rx_ring_num; i++) {
1413 		cq = priv->rx_cq[i];
1414 		cq->moder_cnt = priv->rx_frames;
1415 		cq->moder_time = priv->rx_usecs;
1416 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1417 		priv->last_moder_packets[i] = 0;
1418 		priv->last_moder_bytes[i] = 0;
1419 	}
1420 
1421 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1422 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1423 			cq = priv->tx_cq[t][i];
1424 			cq->moder_cnt = priv->tx_frames;
1425 			cq->moder_time = priv->tx_usecs;
1426 		}
1427 	}
1428 
1429 	/* Reset auto-moderation params */
1430 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1431 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1432 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1433 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1434 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1435 	priv->adaptive_rx_coal = 1;
1436 	priv->last_moder_jiffies = 0;
1437 	priv->last_moder_tx_packets = 0;
1438 }
1439 
1440 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1441 {
1442 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1443 	u32 pkt_rate_high, pkt_rate_low;
1444 	struct mlx4_en_cq *cq;
1445 	unsigned long packets;
1446 	unsigned long rate;
1447 	unsigned long avg_pkt_size;
1448 	unsigned long rx_packets;
1449 	unsigned long rx_bytes;
1450 	unsigned long rx_pkt_diff;
1451 	int moder_time;
1452 	int ring, err;
1453 
1454 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1455 		return;
1456 
1457 	pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1458 	pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1459 
1460 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1461 		rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1462 		rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1463 
1464 		rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1465 		packets = rx_pkt_diff;
1466 		rate = packets * HZ / period;
1467 		avg_pkt_size = packets ? (rx_bytes -
1468 				priv->last_moder_bytes[ring]) / packets : 0;
1469 
1470 		/* Apply auto-moderation only when packet rate
1471 		 * exceeds a rate that it matters */
1472 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1473 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1474 			if (rate <= pkt_rate_low)
1475 				moder_time = priv->rx_usecs_low;
1476 			else if (rate >= pkt_rate_high)
1477 				moder_time = priv->rx_usecs_high;
1478 			else
1479 				moder_time = (rate - pkt_rate_low) *
1480 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1481 					(pkt_rate_high - pkt_rate_low) +
1482 					priv->rx_usecs_low;
1483 		} else {
1484 			moder_time = priv->rx_usecs_low;
1485 		}
1486 
1487 		cq = priv->rx_cq[ring];
1488 		if (moder_time != priv->last_moder_time[ring] ||
1489 		    cq->moder_cnt != priv->rx_frames) {
1490 			priv->last_moder_time[ring] = moder_time;
1491 			cq->moder_time = moder_time;
1492 			cq->moder_cnt = priv->rx_frames;
1493 			err = mlx4_en_set_cq_moder(priv, cq);
1494 			if (err)
1495 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1496 				       ring);
1497 		}
1498 		priv->last_moder_packets[ring] = rx_packets;
1499 		priv->last_moder_bytes[ring] = rx_bytes;
1500 	}
1501 
1502 	priv->last_moder_jiffies = jiffies;
1503 }
1504 
1505 static void mlx4_en_do_get_stats(struct work_struct *work)
1506 {
1507 	struct delayed_work *delay = to_delayed_work(work);
1508 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1509 						 stats_task);
1510 	struct mlx4_en_dev *mdev = priv->mdev;
1511 	int err;
1512 
1513 	mutex_lock(&mdev->state_lock);
1514 	if (mdev->device_up) {
1515 		if (priv->port_up) {
1516 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1517 			if (err)
1518 				en_dbg(HW, priv, "Could not update stats\n");
1519 
1520 			mlx4_en_auto_moderation(priv);
1521 		}
1522 
1523 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1524 	}
1525 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1526 		mlx4_en_do_set_mac(priv, priv->current_mac);
1527 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1528 	}
1529 	mutex_unlock(&mdev->state_lock);
1530 }
1531 
1532 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1533  * periodically
1534  */
1535 static void mlx4_en_service_task(struct work_struct *work)
1536 {
1537 	struct delayed_work *delay = to_delayed_work(work);
1538 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1539 						 service_task);
1540 	struct mlx4_en_dev *mdev = priv->mdev;
1541 
1542 	mutex_lock(&mdev->state_lock);
1543 	if (mdev->device_up) {
1544 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1545 			mlx4_en_ptp_overflow_check(mdev);
1546 
1547 		mlx4_en_recover_from_oom(priv);
1548 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1549 				   SERVICE_TASK_DELAY);
1550 	}
1551 	mutex_unlock(&mdev->state_lock);
1552 }
1553 
1554 static void mlx4_en_linkstate(struct work_struct *work)
1555 {
1556 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1557 						 linkstate_task);
1558 	struct mlx4_en_dev *mdev = priv->mdev;
1559 	int linkstate = priv->link_state;
1560 
1561 	mutex_lock(&mdev->state_lock);
1562 	/* If observable port state changed set carrier state and
1563 	 * report to system log */
1564 	if (priv->last_link_state != linkstate) {
1565 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1566 			en_info(priv, "Link Down\n");
1567 			netif_carrier_off(priv->dev);
1568 		} else {
1569 			en_info(priv, "Link Up\n");
1570 			netif_carrier_on(priv->dev);
1571 		}
1572 	}
1573 	priv->last_link_state = linkstate;
1574 	mutex_unlock(&mdev->state_lock);
1575 }
1576 
1577 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1578 {
1579 	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1580 	int numa_node = priv->mdev->dev->numa_node;
1581 
1582 	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1583 		return -ENOMEM;
1584 
1585 	cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1586 			ring->affinity_mask);
1587 	return 0;
1588 }
1589 
1590 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1591 {
1592 	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1593 }
1594 
1595 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1596 				      int tx_ring_idx)
1597 {
1598 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1599 	int rr_index = tx_ring_idx;
1600 
1601 	tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1602 	tx_ring->recycle_ring = priv->rx_ring[rr_index];
1603 	en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1604 	       TX_XDP, tx_ring_idx, rr_index);
1605 }
1606 
1607 int mlx4_en_start_port(struct net_device *dev)
1608 {
1609 	struct mlx4_en_priv *priv = netdev_priv(dev);
1610 	struct mlx4_en_dev *mdev = priv->mdev;
1611 	struct mlx4_en_cq *cq;
1612 	struct mlx4_en_tx_ring *tx_ring;
1613 	int rx_index = 0;
1614 	int err = 0;
1615 	int i, t;
1616 	int j;
1617 	u8 mc_list[16] = {0};
1618 
1619 	if (priv->port_up) {
1620 		en_dbg(DRV, priv, "start port called while port already up\n");
1621 		return 0;
1622 	}
1623 
1624 	INIT_LIST_HEAD(&priv->mc_list);
1625 	INIT_LIST_HEAD(&priv->curr_list);
1626 	INIT_LIST_HEAD(&priv->ethtool_list);
1627 	memset(&priv->ethtool_rules[0], 0,
1628 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1629 
1630 	/* Calculate Rx buf size */
1631 	dev->mtu = min(dev->mtu, priv->max_mtu);
1632 	mlx4_en_calc_rx_buf(dev);
1633 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1634 
1635 	/* Configure rx cq's and rings */
1636 	err = mlx4_en_activate_rx_rings(priv);
1637 	if (err) {
1638 		en_err(priv, "Failed to activate RX rings\n");
1639 		return err;
1640 	}
1641 	for (i = 0; i < priv->rx_ring_num; i++) {
1642 		cq = priv->rx_cq[i];
1643 
1644 		err = mlx4_en_init_affinity_hint(priv, i);
1645 		if (err) {
1646 			en_err(priv, "Failed preparing IRQ affinity hint\n");
1647 			goto cq_err;
1648 		}
1649 
1650 		err = mlx4_en_activate_cq(priv, cq, i);
1651 		if (err) {
1652 			en_err(priv, "Failed activating Rx CQ\n");
1653 			mlx4_en_free_affinity_hint(priv, i);
1654 			goto cq_err;
1655 		}
1656 
1657 		for (j = 0; j < cq->size; j++) {
1658 			struct mlx4_cqe *cqe = NULL;
1659 
1660 			cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1661 			      priv->cqe_factor;
1662 			cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1663 		}
1664 
1665 		err = mlx4_en_set_cq_moder(priv, cq);
1666 		if (err) {
1667 			en_err(priv, "Failed setting cq moderation parameters\n");
1668 			mlx4_en_deactivate_cq(priv, cq);
1669 			mlx4_en_free_affinity_hint(priv, i);
1670 			goto cq_err;
1671 		}
1672 		mlx4_en_arm_cq(priv, cq);
1673 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1674 		++rx_index;
1675 	}
1676 
1677 	/* Set qp number */
1678 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1679 	err = mlx4_en_get_qp(priv);
1680 	if (err) {
1681 		en_err(priv, "Failed getting eth qp\n");
1682 		goto cq_err;
1683 	}
1684 	mdev->mac_removed[priv->port] = 0;
1685 
1686 	priv->counter_index =
1687 			mlx4_get_default_counter_index(mdev->dev, priv->port);
1688 
1689 	err = mlx4_en_config_rss_steer(priv);
1690 	if (err) {
1691 		en_err(priv, "Failed configuring rss steering\n");
1692 		goto mac_err;
1693 	}
1694 
1695 	err = mlx4_en_create_drop_qp(priv);
1696 	if (err)
1697 		goto rss_err;
1698 
1699 	/* Configure tx cq's and rings */
1700 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1701 		u8 num_tx_rings_p_up = t == TX ?
1702 			priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1703 
1704 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1705 			/* Configure cq */
1706 			cq = priv->tx_cq[t][i];
1707 			err = mlx4_en_activate_cq(priv, cq, i);
1708 			if (err) {
1709 				en_err(priv, "Failed allocating Tx CQ\n");
1710 				goto tx_err;
1711 			}
1712 			err = mlx4_en_set_cq_moder(priv, cq);
1713 			if (err) {
1714 				en_err(priv, "Failed setting cq moderation parameters\n");
1715 				mlx4_en_deactivate_cq(priv, cq);
1716 				goto tx_err;
1717 			}
1718 			en_dbg(DRV, priv,
1719 			       "Resetting index of collapsed CQ:%d to -1\n", i);
1720 			cq->buf->wqe_index = cpu_to_be16(0xffff);
1721 
1722 			/* Configure ring */
1723 			tx_ring = priv->tx_ring[t][i];
1724 			err = mlx4_en_activate_tx_ring(priv, tx_ring,
1725 						       cq->mcq.cqn,
1726 						       i / num_tx_rings_p_up);
1727 			if (err) {
1728 				en_err(priv, "Failed allocating Tx ring\n");
1729 				mlx4_en_deactivate_cq(priv, cq);
1730 				goto tx_err;
1731 			}
1732 			if (t != TX_XDP) {
1733 				tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1734 				tx_ring->recycle_ring = NULL;
1735 
1736 				/* Arm CQ for TX completions */
1737 				mlx4_en_arm_cq(priv, cq);
1738 
1739 			} else {
1740 				mlx4_en_init_tx_xdp_ring_descs(priv, tx_ring);
1741 				mlx4_en_init_recycle_ring(priv, i);
1742 				/* XDP TX CQ should never be armed */
1743 			}
1744 
1745 			/* Set initial ownership of all Tx TXBBs to SW (1) */
1746 			for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1747 				*((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1748 		}
1749 	}
1750 
1751 	/* Configure port */
1752 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1753 				    priv->rx_skb_size + ETH_FCS_LEN,
1754 				    priv->prof->tx_pause,
1755 				    priv->prof->tx_ppp,
1756 				    priv->prof->rx_pause,
1757 				    priv->prof->rx_ppp);
1758 	if (err) {
1759 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1760 		       priv->port, err);
1761 		goto tx_err;
1762 	}
1763 
1764 	err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1765 	if (err) {
1766 		en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1767 		       dev->mtu, priv->port, err);
1768 		goto tx_err;
1769 	}
1770 
1771 	/* Set default qp number */
1772 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1773 	if (err) {
1774 		en_err(priv, "Failed setting default qp numbers\n");
1775 		goto tx_err;
1776 	}
1777 
1778 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1779 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1780 		if (err) {
1781 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1782 			       err);
1783 			goto tx_err;
1784 		}
1785 	}
1786 
1787 	/* Init port */
1788 	en_dbg(HW, priv, "Initializing port\n");
1789 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1790 	if (err) {
1791 		en_err(priv, "Failed Initializing port\n");
1792 		goto tx_err;
1793 	}
1794 
1795 	/* Set Unicast and VXLAN steering rules */
1796 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1797 	    mlx4_en_set_rss_steer_rules(priv))
1798 		mlx4_warn(mdev, "Failed setting steering rules\n");
1799 
1800 	/* Attach rx QP to bradcast address */
1801 	eth_broadcast_addr(&mc_list[10]);
1802 	mc_list[5] = priv->port; /* needed for B0 steering support */
1803 	if (mlx4_multicast_attach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1804 				  priv->port, 0, MLX4_PROT_ETH,
1805 				  &priv->broadcast_id))
1806 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1807 
1808 	/* Must redo promiscuous mode setup. */
1809 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1810 
1811 	/* Schedule multicast task to populate multicast list */
1812 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1813 
1814 	if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1815 		udp_tunnel_get_rx_info(dev);
1816 
1817 	priv->port_up = true;
1818 
1819 	/* Process all completions if exist to prevent
1820 	 * the queues freezing if they are full
1821 	 */
1822 	for (i = 0; i < priv->rx_ring_num; i++) {
1823 		local_bh_disable();
1824 		napi_schedule(&priv->rx_cq[i]->napi);
1825 		local_bh_enable();
1826 	}
1827 
1828 	netif_tx_start_all_queues(dev);
1829 	netif_device_attach(dev);
1830 
1831 	return 0;
1832 
1833 tx_err:
1834 	if (t == MLX4_EN_NUM_TX_TYPES) {
1835 		t--;
1836 		i = priv->tx_ring_num[t];
1837 	}
1838 	while (t >= 0) {
1839 		while (i--) {
1840 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1841 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1842 		}
1843 		if (!t--)
1844 			break;
1845 		i = priv->tx_ring_num[t];
1846 	}
1847 	mlx4_en_destroy_drop_qp(priv);
1848 rss_err:
1849 	mlx4_en_release_rss_steer(priv);
1850 mac_err:
1851 	mlx4_en_put_qp(priv);
1852 cq_err:
1853 	while (rx_index--) {
1854 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1855 		mlx4_en_free_affinity_hint(priv, rx_index);
1856 	}
1857 	for (i = 0; i < priv->rx_ring_num; i++)
1858 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1859 
1860 	return err; /* need to close devices */
1861 }
1862 
1863 
1864 void mlx4_en_stop_port(struct net_device *dev, int detach)
1865 {
1866 	struct mlx4_en_priv *priv = netdev_priv(dev);
1867 	struct mlx4_en_dev *mdev = priv->mdev;
1868 	struct mlx4_en_mc_list *mclist, *tmp;
1869 	struct ethtool_flow_id *flow, *tmp_flow;
1870 	int i, t;
1871 	u8 mc_list[16] = {0};
1872 
1873 	if (!priv->port_up) {
1874 		en_dbg(DRV, priv, "stop port called while port already down\n");
1875 		return;
1876 	}
1877 
1878 	/* close port*/
1879 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1880 
1881 	/* Synchronize with tx routine */
1882 	netif_tx_lock_bh(dev);
1883 	if (detach)
1884 		netif_device_detach(dev);
1885 	netif_tx_stop_all_queues(dev);
1886 	netif_tx_unlock_bh(dev);
1887 
1888 	netif_tx_disable(dev);
1889 
1890 	spin_lock_bh(&priv->stats_lock);
1891 	mlx4_en_fold_software_stats(dev);
1892 	/* Set port as not active */
1893 	priv->port_up = false;
1894 	spin_unlock_bh(&priv->stats_lock);
1895 
1896 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1897 
1898 	/* Promsicuous mode */
1899 	if (mdev->dev->caps.steering_mode ==
1900 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1901 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1902 				 MLX4_EN_FLAG_MC_PROMISC);
1903 		mlx4_flow_steer_promisc_remove(mdev->dev,
1904 					       priv->port,
1905 					       MLX4_FS_ALL_DEFAULT);
1906 		mlx4_flow_steer_promisc_remove(mdev->dev,
1907 					       priv->port,
1908 					       MLX4_FS_MC_DEFAULT);
1909 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1910 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1911 
1912 		/* Disable promiscouos mode */
1913 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1914 					    priv->port);
1915 
1916 		/* Disable Multicast promisc */
1917 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1918 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1919 						      priv->port);
1920 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1921 		}
1922 	}
1923 
1924 	/* Detach All multicasts */
1925 	eth_broadcast_addr(&mc_list[10]);
1926 	mc_list[5] = priv->port; /* needed for B0 steering support */
1927 	mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1928 			      MLX4_PROT_ETH, priv->broadcast_id);
1929 	list_for_each_entry(mclist, &priv->curr_list, list) {
1930 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1931 		mc_list[5] = priv->port;
1932 		mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp,
1933 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1934 		if (mclist->tunnel_reg_id)
1935 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1936 	}
1937 	mlx4_en_clear_list(dev);
1938 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1939 		list_del(&mclist->list);
1940 		kfree(mclist);
1941 	}
1942 
1943 	/* Flush multicast filter */
1944 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1945 
1946 	/* Remove flow steering rules for the port*/
1947 	if (mdev->dev->caps.steering_mode ==
1948 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1949 		ASSERT_RTNL();
1950 		list_for_each_entry_safe(flow, tmp_flow,
1951 					 &priv->ethtool_list, list) {
1952 			mlx4_flow_detach(mdev->dev, flow->id);
1953 			list_del(&flow->list);
1954 		}
1955 	}
1956 
1957 	mlx4_en_destroy_drop_qp(priv);
1958 
1959 	/* Free TX Rings */
1960 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1961 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1962 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1963 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1964 		}
1965 	}
1966 	msleep(10);
1967 
1968 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1969 		for (i = 0; i < priv->tx_ring_num[t]; i++)
1970 			mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1971 
1972 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1973 		mlx4_en_delete_rss_steer_rules(priv);
1974 
1975 	/* Free RSS qps */
1976 	mlx4_en_release_rss_steer(priv);
1977 
1978 	/* Unregister Mac address for the port */
1979 	mlx4_en_put_qp(priv);
1980 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1981 		mdev->mac_removed[priv->port] = 1;
1982 
1983 	/* Free RX Rings */
1984 	for (i = 0; i < priv->rx_ring_num; i++) {
1985 		struct mlx4_en_cq *cq = priv->rx_cq[i];
1986 
1987 		napi_synchronize(&cq->napi);
1988 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1989 		mlx4_en_deactivate_cq(priv, cq);
1990 
1991 		mlx4_en_free_affinity_hint(priv, i);
1992 	}
1993 }
1994 
1995 static void mlx4_en_restart(struct work_struct *work)
1996 {
1997 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1998 						 watchdog_task);
1999 	struct mlx4_en_dev *mdev = priv->mdev;
2000 	struct net_device *dev = priv->dev;
2001 
2002 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
2003 
2004 	rtnl_lock();
2005 	mutex_lock(&mdev->state_lock);
2006 	if (priv->port_up) {
2007 		mlx4_en_stop_port(dev, 1);
2008 		if (mlx4_en_start_port(dev))
2009 			en_err(priv, "Failed restarting port %d\n", priv->port);
2010 	}
2011 	mutex_unlock(&mdev->state_lock);
2012 	rtnl_unlock();
2013 }
2014 
2015 static void mlx4_en_clear_stats(struct net_device *dev)
2016 {
2017 	struct mlx4_en_priv *priv = netdev_priv(dev);
2018 	struct mlx4_en_dev *mdev = priv->mdev;
2019 	struct mlx4_en_tx_ring **tx_ring;
2020 	int i;
2021 
2022 	if (!mlx4_is_slave(mdev->dev))
2023 		if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
2024 			en_dbg(HW, priv, "Failed dumping statistics\n");
2025 
2026 	memset(&priv->pstats, 0, sizeof(priv->pstats));
2027 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
2028 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
2029 	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
2030 	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
2031 	memset(&priv->rx_priority_flowstats, 0,
2032 	       sizeof(priv->rx_priority_flowstats));
2033 	memset(&priv->tx_priority_flowstats, 0,
2034 	       sizeof(priv->tx_priority_flowstats));
2035 	memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
2036 
2037 	tx_ring = priv->tx_ring[TX];
2038 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
2039 		tx_ring[i]->bytes = 0;
2040 		tx_ring[i]->packets = 0;
2041 		tx_ring[i]->tx_csum = 0;
2042 		tx_ring[i]->tx_dropped = 0;
2043 		tx_ring[i]->queue_stopped = 0;
2044 		tx_ring[i]->wake_queue = 0;
2045 		tx_ring[i]->tso_packets = 0;
2046 		tx_ring[i]->xmit_more = 0;
2047 	}
2048 	for (i = 0; i < priv->rx_ring_num; i++) {
2049 		priv->rx_ring[i]->bytes = 0;
2050 		priv->rx_ring[i]->packets = 0;
2051 		priv->rx_ring[i]->csum_ok = 0;
2052 		priv->rx_ring[i]->csum_none = 0;
2053 		priv->rx_ring[i]->csum_complete = 0;
2054 	}
2055 }
2056 
2057 static int mlx4_en_open(struct net_device *dev)
2058 {
2059 	struct mlx4_en_priv *priv = netdev_priv(dev);
2060 	struct mlx4_en_dev *mdev = priv->mdev;
2061 	int err = 0;
2062 
2063 	mutex_lock(&mdev->state_lock);
2064 
2065 	if (!mdev->device_up) {
2066 		en_err(priv, "Cannot open - device down/disabled\n");
2067 		err = -EBUSY;
2068 		goto out;
2069 	}
2070 
2071 	/* Reset HW statistics and SW counters */
2072 	mlx4_en_clear_stats(dev);
2073 
2074 	err = mlx4_en_start_port(dev);
2075 	if (err)
2076 		en_err(priv, "Failed starting port:%d\n", priv->port);
2077 
2078 out:
2079 	mutex_unlock(&mdev->state_lock);
2080 	return err;
2081 }
2082 
2083 
2084 static int mlx4_en_close(struct net_device *dev)
2085 {
2086 	struct mlx4_en_priv *priv = netdev_priv(dev);
2087 	struct mlx4_en_dev *mdev = priv->mdev;
2088 
2089 	en_dbg(IFDOWN, priv, "Close port called\n");
2090 
2091 	mutex_lock(&mdev->state_lock);
2092 
2093 	mlx4_en_stop_port(dev, 0);
2094 	netif_carrier_off(dev);
2095 
2096 	mutex_unlock(&mdev->state_lock);
2097 	return 0;
2098 }
2099 
2100 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2101 {
2102 	int i, t;
2103 
2104 #ifdef CONFIG_RFS_ACCEL
2105 	priv->dev->rx_cpu_rmap = NULL;
2106 #endif
2107 
2108 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2109 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2110 			if (priv->tx_ring[t] && priv->tx_ring[t][i])
2111 				mlx4_en_destroy_tx_ring(priv,
2112 							&priv->tx_ring[t][i]);
2113 			if (priv->tx_cq[t] && priv->tx_cq[t][i])
2114 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2115 		}
2116 		kfree(priv->tx_ring[t]);
2117 		kfree(priv->tx_cq[t]);
2118 	}
2119 
2120 	for (i = 0; i < priv->rx_ring_num; i++) {
2121 		if (priv->rx_ring[i])
2122 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2123 				priv->prof->rx_ring_size, priv->stride);
2124 		if (priv->rx_cq[i])
2125 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2126 	}
2127 
2128 }
2129 
2130 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2131 {
2132 	struct mlx4_en_port_profile *prof = priv->prof;
2133 	int i, t;
2134 	int node;
2135 
2136 	/* Create tx Rings */
2137 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2138 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2139 			node = cpu_to_node(i % num_online_cpus());
2140 			if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2141 					      prof->tx_ring_size, i, t, node))
2142 				goto err;
2143 
2144 			if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2145 						   prof->tx_ring_size,
2146 						   TXBB_SIZE, node, i))
2147 				goto err;
2148 		}
2149 	}
2150 
2151 	/* Create rx Rings */
2152 	for (i = 0; i < priv->rx_ring_num; i++) {
2153 		node = cpu_to_node(i % num_online_cpus());
2154 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2155 				      prof->rx_ring_size, i, RX, node))
2156 			goto err;
2157 
2158 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2159 					   prof->rx_ring_size, priv->stride,
2160 					   node, i))
2161 			goto err;
2162 
2163 	}
2164 
2165 #ifdef CONFIG_RFS_ACCEL
2166 	priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2167 #endif
2168 
2169 	return 0;
2170 
2171 err:
2172 	en_err(priv, "Failed to allocate NIC resources\n");
2173 	for (i = 0; i < priv->rx_ring_num; i++) {
2174 		if (priv->rx_ring[i])
2175 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2176 						prof->rx_ring_size,
2177 						priv->stride);
2178 		if (priv->rx_cq[i])
2179 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2180 	}
2181 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2182 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2183 			if (priv->tx_ring[t][i])
2184 				mlx4_en_destroy_tx_ring(priv,
2185 							&priv->tx_ring[t][i]);
2186 			if (priv->tx_cq[t][i])
2187 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2188 		}
2189 	}
2190 	return -ENOMEM;
2191 }
2192 
2193 
2194 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2195 			     struct mlx4_en_priv *src,
2196 			     struct mlx4_en_port_profile *prof)
2197 {
2198 	int t;
2199 
2200 	memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2201 	       sizeof(dst->hwtstamp_config));
2202 	dst->num_tx_rings_p_up = prof->num_tx_rings_p_up;
2203 	dst->rx_ring_num = prof->rx_ring_num;
2204 	dst->flags = prof->flags;
2205 	dst->mdev = src->mdev;
2206 	dst->port = src->port;
2207 	dst->dev = src->dev;
2208 	dst->prof = prof;
2209 	dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2210 					 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2211 
2212 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2213 		dst->tx_ring_num[t] = prof->tx_ring_num[t];
2214 		if (!dst->tx_ring_num[t])
2215 			continue;
2216 
2217 		dst->tx_ring[t] = kcalloc(MAX_TX_RINGS,
2218 					  sizeof(struct mlx4_en_tx_ring *),
2219 					  GFP_KERNEL);
2220 		if (!dst->tx_ring[t])
2221 			goto err_free_tx;
2222 
2223 		dst->tx_cq[t] = kcalloc(MAX_TX_RINGS,
2224 					sizeof(struct mlx4_en_cq *),
2225 					GFP_KERNEL);
2226 		if (!dst->tx_cq[t]) {
2227 			kfree(dst->tx_ring[t]);
2228 			goto err_free_tx;
2229 		}
2230 	}
2231 
2232 	return 0;
2233 
2234 err_free_tx:
2235 	while (t--) {
2236 		kfree(dst->tx_ring[t]);
2237 		kfree(dst->tx_cq[t]);
2238 	}
2239 	return -ENOMEM;
2240 }
2241 
2242 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2243 				struct mlx4_en_priv *src)
2244 {
2245 	int t;
2246 	memcpy(dst->rx_ring, src->rx_ring,
2247 	       sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2248 	memcpy(dst->rx_cq, src->rx_cq,
2249 	       sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2250 	memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2251 	       sizeof(dst->hwtstamp_config));
2252 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2253 		dst->tx_ring_num[t] = src->tx_ring_num[t];
2254 		dst->tx_ring[t] = src->tx_ring[t];
2255 		dst->tx_cq[t] = src->tx_cq[t];
2256 	}
2257 	dst->num_tx_rings_p_up = src->num_tx_rings_p_up;
2258 	dst->rx_ring_num = src->rx_ring_num;
2259 	memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2260 }
2261 
2262 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2263 				struct mlx4_en_priv *tmp,
2264 				struct mlx4_en_port_profile *prof,
2265 				bool carry_xdp_prog)
2266 {
2267 	struct bpf_prog *xdp_prog;
2268 	int i, t;
2269 
2270 	mlx4_en_copy_priv(tmp, priv, prof);
2271 
2272 	if (mlx4_en_alloc_resources(tmp)) {
2273 		en_warn(priv,
2274 			"%s: Resource allocation failed, using previous configuration\n",
2275 			__func__);
2276 		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2277 			kfree(tmp->tx_ring[t]);
2278 			kfree(tmp->tx_cq[t]);
2279 		}
2280 		return -ENOMEM;
2281 	}
2282 
2283 	/* All rx_rings has the same xdp_prog.  Pick the first one. */
2284 	xdp_prog = rcu_dereference_protected(
2285 		priv->rx_ring[0]->xdp_prog,
2286 		lockdep_is_held(&priv->mdev->state_lock));
2287 
2288 	if (xdp_prog && carry_xdp_prog) {
2289 		xdp_prog = bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2290 		if (IS_ERR(xdp_prog)) {
2291 			mlx4_en_free_resources(tmp);
2292 			return PTR_ERR(xdp_prog);
2293 		}
2294 		for (i = 0; i < tmp->rx_ring_num; i++)
2295 			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2296 					   xdp_prog);
2297 	}
2298 
2299 	return 0;
2300 }
2301 
2302 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2303 				    struct mlx4_en_priv *tmp)
2304 {
2305 	mlx4_en_free_resources(priv);
2306 	mlx4_en_update_priv(priv, tmp);
2307 }
2308 
2309 void mlx4_en_destroy_netdev(struct net_device *dev)
2310 {
2311 	struct mlx4_en_priv *priv = netdev_priv(dev);
2312 	struct mlx4_en_dev *mdev = priv->mdev;
2313 
2314 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2315 
2316 	/* Unregister device - this will close the port if it was up */
2317 	if (priv->registered) {
2318 		devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2319 							      priv->port));
2320 		unregister_netdev(dev);
2321 	}
2322 
2323 	if (priv->allocated)
2324 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2325 
2326 	cancel_delayed_work(&priv->stats_task);
2327 	cancel_delayed_work(&priv->service_task);
2328 	/* flush any pending task for this netdev */
2329 	flush_workqueue(mdev->workqueue);
2330 
2331 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2332 		mlx4_en_remove_timestamp(mdev);
2333 
2334 	/* Detach the netdev so tasks would not attempt to access it */
2335 	mutex_lock(&mdev->state_lock);
2336 	mdev->pndev[priv->port] = NULL;
2337 	mdev->upper[priv->port] = NULL;
2338 
2339 #ifdef CONFIG_RFS_ACCEL
2340 	mlx4_en_cleanup_filters(priv);
2341 #endif
2342 
2343 	mlx4_en_free_resources(priv);
2344 	mutex_unlock(&mdev->state_lock);
2345 
2346 	free_netdev(dev);
2347 }
2348 
2349 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2350 {
2351 	struct mlx4_en_priv *priv = netdev_priv(dev);
2352 
2353 	if (mtu > MLX4_EN_MAX_XDP_MTU) {
2354 		en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2355 		       mtu, MLX4_EN_MAX_XDP_MTU);
2356 		return false;
2357 	}
2358 
2359 	return true;
2360 }
2361 
2362 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2363 {
2364 	struct mlx4_en_priv *priv = netdev_priv(dev);
2365 	struct mlx4_en_dev *mdev = priv->mdev;
2366 	int err = 0;
2367 
2368 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2369 		 dev->mtu, new_mtu);
2370 
2371 	if (priv->tx_ring_num[TX_XDP] &&
2372 	    !mlx4_en_check_xdp_mtu(dev, new_mtu))
2373 		return -EOPNOTSUPP;
2374 
2375 	dev->mtu = new_mtu;
2376 
2377 	if (netif_running(dev)) {
2378 		mutex_lock(&mdev->state_lock);
2379 		if (!mdev->device_up) {
2380 			/* NIC is probably restarting - let watchdog task reset
2381 			 * the port */
2382 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2383 		} else {
2384 			mlx4_en_stop_port(dev, 1);
2385 			err = mlx4_en_start_port(dev);
2386 			if (err) {
2387 				en_err(priv, "Failed restarting port:%d\n",
2388 					 priv->port);
2389 				queue_work(mdev->workqueue, &priv->watchdog_task);
2390 			}
2391 		}
2392 		mutex_unlock(&mdev->state_lock);
2393 	}
2394 	return 0;
2395 }
2396 
2397 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2398 {
2399 	struct mlx4_en_priv *priv = netdev_priv(dev);
2400 	struct mlx4_en_dev *mdev = priv->mdev;
2401 	struct hwtstamp_config config;
2402 
2403 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2404 		return -EFAULT;
2405 
2406 	/* reserved for future extensions */
2407 	if (config.flags)
2408 		return -EINVAL;
2409 
2410 	/* device doesn't support time stamping */
2411 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2412 		return -EINVAL;
2413 
2414 	/* TX HW timestamp */
2415 	switch (config.tx_type) {
2416 	case HWTSTAMP_TX_OFF:
2417 	case HWTSTAMP_TX_ON:
2418 		break;
2419 	default:
2420 		return -ERANGE;
2421 	}
2422 
2423 	/* RX HW timestamp */
2424 	switch (config.rx_filter) {
2425 	case HWTSTAMP_FILTER_NONE:
2426 		break;
2427 	case HWTSTAMP_FILTER_ALL:
2428 	case HWTSTAMP_FILTER_SOME:
2429 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2430 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2431 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2432 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2433 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2434 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2435 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2436 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2437 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2438 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2439 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2440 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2441 	case HWTSTAMP_FILTER_NTP_ALL:
2442 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2443 		break;
2444 	default:
2445 		return -ERANGE;
2446 	}
2447 
2448 	if (mlx4_en_reset_config(dev, config, dev->features)) {
2449 		config.tx_type = HWTSTAMP_TX_OFF;
2450 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2451 	}
2452 
2453 	return copy_to_user(ifr->ifr_data, &config,
2454 			    sizeof(config)) ? -EFAULT : 0;
2455 }
2456 
2457 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2458 {
2459 	struct mlx4_en_priv *priv = netdev_priv(dev);
2460 
2461 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2462 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2463 }
2464 
2465 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2466 {
2467 	switch (cmd) {
2468 	case SIOCSHWTSTAMP:
2469 		return mlx4_en_hwtstamp_set(dev, ifr);
2470 	case SIOCGHWTSTAMP:
2471 		return mlx4_en_hwtstamp_get(dev, ifr);
2472 	default:
2473 		return -EOPNOTSUPP;
2474 	}
2475 }
2476 
2477 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2478 					      netdev_features_t features)
2479 {
2480 	struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2481 	struct mlx4_en_dev *mdev = en_priv->mdev;
2482 
2483 	/* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2484 	 * enable/disable make sure S-TAG flag is always in same state as
2485 	 * C-TAG.
2486 	 */
2487 	if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2488 	    !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2489 		features |= NETIF_F_HW_VLAN_STAG_RX;
2490 	else
2491 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2492 
2493 	return features;
2494 }
2495 
2496 static int mlx4_en_set_features(struct net_device *netdev,
2497 		netdev_features_t features)
2498 {
2499 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2500 	bool reset = false;
2501 	int ret = 0;
2502 
2503 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2504 		en_info(priv, "Turn %s RX-FCS\n",
2505 			(features & NETIF_F_RXFCS) ? "ON" : "OFF");
2506 		reset = true;
2507 	}
2508 
2509 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2510 		u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2511 
2512 		en_info(priv, "Turn %s RX-ALL\n",
2513 			ignore_fcs_value ? "ON" : "OFF");
2514 		ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2515 					      priv->port, ignore_fcs_value);
2516 		if (ret)
2517 			return ret;
2518 	}
2519 
2520 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2521 		en_info(priv, "Turn %s RX vlan strip offload\n",
2522 			(features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2523 		reset = true;
2524 	}
2525 
2526 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2527 		en_info(priv, "Turn %s TX vlan strip offload\n",
2528 			(features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2529 
2530 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2531 		en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2532 			(features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2533 
2534 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2535 		en_info(priv, "Turn %s loopback\n",
2536 			(features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2537 		mlx4_en_update_loopback_state(netdev, features);
2538 	}
2539 
2540 	if (reset) {
2541 		ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2542 					   features);
2543 		if (ret)
2544 			return ret;
2545 	}
2546 
2547 	return 0;
2548 }
2549 
2550 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2551 {
2552 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2553 	struct mlx4_en_dev *mdev = en_priv->mdev;
2554 
2555 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac);
2556 }
2557 
2558 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2559 			       __be16 vlan_proto)
2560 {
2561 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2562 	struct mlx4_en_dev *mdev = en_priv->mdev;
2563 
2564 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2565 				vlan_proto);
2566 }
2567 
2568 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2569 			       int max_tx_rate)
2570 {
2571 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2572 	struct mlx4_en_dev *mdev = en_priv->mdev;
2573 
2574 	return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2575 				max_tx_rate);
2576 }
2577 
2578 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2579 {
2580 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2581 	struct mlx4_en_dev *mdev = en_priv->mdev;
2582 
2583 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2584 }
2585 
2586 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2587 {
2588 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2589 	struct mlx4_en_dev *mdev = en_priv->mdev;
2590 
2591 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2592 }
2593 
2594 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2595 {
2596 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2597 	struct mlx4_en_dev *mdev = en_priv->mdev;
2598 
2599 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2600 }
2601 
2602 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2603 				struct ifla_vf_stats *vf_stats)
2604 {
2605 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2606 	struct mlx4_en_dev *mdev = en_priv->mdev;
2607 
2608 	return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2609 }
2610 
2611 #define PORT_ID_BYTE_LEN 8
2612 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2613 				    struct netdev_phys_item_id *ppid)
2614 {
2615 	struct mlx4_en_priv *priv = netdev_priv(dev);
2616 	struct mlx4_dev *mdev = priv->mdev->dev;
2617 	int i;
2618 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2619 
2620 	if (!phys_port_id)
2621 		return -EOPNOTSUPP;
2622 
2623 	ppid->id_len = sizeof(phys_port_id);
2624 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2625 		ppid->id[i] =  phys_port_id & 0xff;
2626 		phys_port_id >>= 8;
2627 	}
2628 	return 0;
2629 }
2630 
2631 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2632 {
2633 	int ret;
2634 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2635 						 vxlan_add_task);
2636 
2637 	ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2638 	if (ret)
2639 		goto out;
2640 
2641 	ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2642 				  VXLAN_STEER_BY_OUTER_MAC, 1);
2643 out:
2644 	if (ret) {
2645 		en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2646 		return;
2647 	}
2648 }
2649 
2650 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2651 {
2652 	int ret;
2653 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2654 						 vxlan_del_task);
2655 	ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2656 				  VXLAN_STEER_BY_OUTER_MAC, 0);
2657 	if (ret)
2658 		en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2659 
2660 	priv->vxlan_port = 0;
2661 }
2662 
2663 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2664 				   struct udp_tunnel_info *ti)
2665 {
2666 	struct mlx4_en_priv *priv = netdev_priv(dev);
2667 	__be16 port = ti->port;
2668 	__be16 current_port;
2669 
2670 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2671 		return;
2672 
2673 	if (ti->sa_family != AF_INET)
2674 		return;
2675 
2676 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2677 		return;
2678 
2679 	current_port = priv->vxlan_port;
2680 	if (current_port && current_port != port) {
2681 		en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2682 			ntohs(current_port), ntohs(port));
2683 		return;
2684 	}
2685 
2686 	priv->vxlan_port = port;
2687 	queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2688 }
2689 
2690 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2691 				   struct udp_tunnel_info *ti)
2692 {
2693 	struct mlx4_en_priv *priv = netdev_priv(dev);
2694 	__be16 port = ti->port;
2695 	__be16 current_port;
2696 
2697 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2698 		return;
2699 
2700 	if (ti->sa_family != AF_INET)
2701 		return;
2702 
2703 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2704 		return;
2705 
2706 	current_port = priv->vxlan_port;
2707 	if (current_port != port) {
2708 		en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2709 		return;
2710 	}
2711 
2712 	queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2713 }
2714 
2715 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2716 						struct net_device *dev,
2717 						netdev_features_t features)
2718 {
2719 	features = vlan_features_check(skb, features);
2720 	features = vxlan_features_check(skb, features);
2721 
2722 	/* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2723 	 * support inner IPv6 checksums and segmentation so  we need to
2724 	 * strip that feature if this is an IPv6 encapsulated frame.
2725 	 */
2726 	if (skb->encapsulation &&
2727 	    (skb->ip_summed == CHECKSUM_PARTIAL)) {
2728 		struct mlx4_en_priv *priv = netdev_priv(dev);
2729 
2730 		if (!priv->vxlan_port ||
2731 		    (ip_hdr(skb)->version != 4) ||
2732 		    (udp_hdr(skb)->dest != priv->vxlan_port))
2733 			features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2734 	}
2735 
2736 	return features;
2737 }
2738 
2739 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2740 {
2741 	struct mlx4_en_priv *priv = netdev_priv(dev);
2742 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2743 	struct mlx4_update_qp_params params;
2744 	int err;
2745 
2746 	if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2747 		return -EOPNOTSUPP;
2748 
2749 	/* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2750 	if (maxrate >> 12) {
2751 		params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2752 		params.rate_val  = maxrate / 1000;
2753 	} else if (maxrate) {
2754 		params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2755 		params.rate_val  = maxrate;
2756 	} else { /* zero serves to revoke the QP rate-limitation */
2757 		params.rate_unit = 0;
2758 		params.rate_val  = 0;
2759 	}
2760 
2761 	err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2762 			     &params);
2763 	return err;
2764 }
2765 
2766 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2767 {
2768 	struct mlx4_en_priv *priv = netdev_priv(dev);
2769 	struct mlx4_en_dev *mdev = priv->mdev;
2770 	struct mlx4_en_port_profile new_prof;
2771 	struct bpf_prog *old_prog;
2772 	struct mlx4_en_priv *tmp;
2773 	int tx_changed = 0;
2774 	int xdp_ring_num;
2775 	int port_up = 0;
2776 	int err;
2777 	int i;
2778 
2779 	xdp_ring_num = prog ? priv->rx_ring_num : 0;
2780 
2781 	/* No need to reconfigure buffers when simply swapping the
2782 	 * program for a new one.
2783 	 */
2784 	if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2785 		if (prog) {
2786 			prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2787 			if (IS_ERR(prog))
2788 				return PTR_ERR(prog);
2789 		}
2790 		mutex_lock(&mdev->state_lock);
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 		mutex_unlock(&mdev->state_lock);
2800 		return 0;
2801 	}
2802 
2803 	if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2804 		return -EOPNOTSUPP;
2805 
2806 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2807 	if (!tmp)
2808 		return -ENOMEM;
2809 
2810 	if (prog) {
2811 		prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2812 		if (IS_ERR(prog)) {
2813 			err = PTR_ERR(prog);
2814 			goto out;
2815 		}
2816 	}
2817 
2818 	mutex_lock(&mdev->state_lock);
2819 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2820 	new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2821 
2822 	if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2823 		tx_changed = 1;
2824 		new_prof.tx_ring_num[TX] =
2825 			MAX_TX_RINGS - ALIGN(xdp_ring_num, priv->prof->num_up);
2826 		en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2827 	}
2828 
2829 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2830 	if (err) {
2831 		if (prog)
2832 			bpf_prog_sub(prog, priv->rx_ring_num - 1);
2833 		goto unlock_out;
2834 	}
2835 
2836 	if (priv->port_up) {
2837 		port_up = 1;
2838 		mlx4_en_stop_port(dev, 1);
2839 	}
2840 
2841 	mlx4_en_safe_replace_resources(priv, tmp);
2842 	if (tx_changed)
2843 		netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2844 
2845 	for (i = 0; i < priv->rx_ring_num; i++) {
2846 		old_prog = rcu_dereference_protected(
2847 					priv->rx_ring[i]->xdp_prog,
2848 					lockdep_is_held(&mdev->state_lock));
2849 		rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2850 		if (old_prog)
2851 			bpf_prog_put(old_prog);
2852 	}
2853 
2854 	if (port_up) {
2855 		err = mlx4_en_start_port(dev);
2856 		if (err) {
2857 			en_err(priv, "Failed starting port %d for XDP change\n",
2858 			       priv->port);
2859 			queue_work(mdev->workqueue, &priv->watchdog_task);
2860 		}
2861 	}
2862 
2863 unlock_out:
2864 	mutex_unlock(&mdev->state_lock);
2865 out:
2866 	kfree(tmp);
2867 	return err;
2868 }
2869 
2870 static u32 mlx4_xdp_query(struct net_device *dev)
2871 {
2872 	struct mlx4_en_priv *priv = netdev_priv(dev);
2873 	struct mlx4_en_dev *mdev = priv->mdev;
2874 	const struct bpf_prog *xdp_prog;
2875 	u32 prog_id = 0;
2876 
2877 	if (!priv->tx_ring_num[TX_XDP])
2878 		return prog_id;
2879 
2880 	mutex_lock(&mdev->state_lock);
2881 	xdp_prog = rcu_dereference_protected(
2882 		priv->rx_ring[0]->xdp_prog,
2883 		lockdep_is_held(&mdev->state_lock));
2884 	if (xdp_prog)
2885 		prog_id = xdp_prog->aux->id;
2886 	mutex_unlock(&mdev->state_lock);
2887 
2888 	return prog_id;
2889 }
2890 
2891 static int mlx4_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2892 {
2893 	switch (xdp->command) {
2894 	case XDP_SETUP_PROG:
2895 		return mlx4_xdp_set(dev, xdp->prog);
2896 	case XDP_QUERY_PROG:
2897 		xdp->prog_id = mlx4_xdp_query(dev);
2898 		return 0;
2899 	default:
2900 		return -EINVAL;
2901 	}
2902 }
2903 
2904 static const struct net_device_ops mlx4_netdev_ops = {
2905 	.ndo_open		= mlx4_en_open,
2906 	.ndo_stop		= mlx4_en_close,
2907 	.ndo_start_xmit		= mlx4_en_xmit,
2908 	.ndo_select_queue	= mlx4_en_select_queue,
2909 	.ndo_get_stats64	= mlx4_en_get_stats64,
2910 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2911 	.ndo_set_mac_address	= mlx4_en_set_mac,
2912 	.ndo_validate_addr	= eth_validate_addr,
2913 	.ndo_change_mtu		= mlx4_en_change_mtu,
2914 	.ndo_do_ioctl		= mlx4_en_ioctl,
2915 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2916 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2917 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2918 	.ndo_set_features	= mlx4_en_set_features,
2919 	.ndo_fix_features	= mlx4_en_fix_features,
2920 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2921 #ifdef CONFIG_RFS_ACCEL
2922 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2923 #endif
2924 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2925 	.ndo_udp_tunnel_add	= mlx4_en_add_vxlan_port,
2926 	.ndo_udp_tunnel_del	= mlx4_en_del_vxlan_port,
2927 	.ndo_features_check	= mlx4_en_features_check,
2928 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2929 	.ndo_bpf		= mlx4_xdp,
2930 };
2931 
2932 static const struct net_device_ops mlx4_netdev_ops_master = {
2933 	.ndo_open		= mlx4_en_open,
2934 	.ndo_stop		= mlx4_en_close,
2935 	.ndo_start_xmit		= mlx4_en_xmit,
2936 	.ndo_select_queue	= mlx4_en_select_queue,
2937 	.ndo_get_stats64	= mlx4_en_get_stats64,
2938 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2939 	.ndo_set_mac_address	= mlx4_en_set_mac,
2940 	.ndo_validate_addr	= eth_validate_addr,
2941 	.ndo_change_mtu		= mlx4_en_change_mtu,
2942 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2943 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2944 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2945 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2946 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2947 	.ndo_set_vf_rate	= mlx4_en_set_vf_rate,
2948 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2949 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2950 	.ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2951 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2952 	.ndo_set_features	= mlx4_en_set_features,
2953 	.ndo_fix_features	= mlx4_en_fix_features,
2954 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2955 #ifdef CONFIG_RFS_ACCEL
2956 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2957 #endif
2958 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2959 	.ndo_udp_tunnel_add	= mlx4_en_add_vxlan_port,
2960 	.ndo_udp_tunnel_del	= mlx4_en_del_vxlan_port,
2961 	.ndo_features_check	= mlx4_en_features_check,
2962 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2963 	.ndo_bpf		= mlx4_xdp,
2964 };
2965 
2966 struct mlx4_en_bond {
2967 	struct work_struct work;
2968 	struct mlx4_en_priv *priv;
2969 	int is_bonded;
2970 	struct mlx4_port_map port_map;
2971 };
2972 
2973 static void mlx4_en_bond_work(struct work_struct *work)
2974 {
2975 	struct mlx4_en_bond *bond = container_of(work,
2976 						     struct mlx4_en_bond,
2977 						     work);
2978 	int err = 0;
2979 	struct mlx4_dev *dev = bond->priv->mdev->dev;
2980 
2981 	if (bond->is_bonded) {
2982 		if (!mlx4_is_bonded(dev)) {
2983 			err = mlx4_bond(dev);
2984 			if (err)
2985 				en_err(bond->priv, "Fail to bond device\n");
2986 		}
2987 		if (!err) {
2988 			err = mlx4_port_map_set(dev, &bond->port_map);
2989 			if (err)
2990 				en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2991 				       bond->port_map.port1,
2992 				       bond->port_map.port2,
2993 				       err);
2994 		}
2995 	} else if (mlx4_is_bonded(dev)) {
2996 		err = mlx4_unbond(dev);
2997 		if (err)
2998 			en_err(bond->priv, "Fail to unbond device\n");
2999 	}
3000 	dev_put(bond->priv->dev);
3001 	kfree(bond);
3002 }
3003 
3004 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
3005 				   u8 v2p_p1, u8 v2p_p2)
3006 {
3007 	struct mlx4_en_bond *bond = NULL;
3008 
3009 	bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
3010 	if (!bond)
3011 		return -ENOMEM;
3012 
3013 	INIT_WORK(&bond->work, mlx4_en_bond_work);
3014 	bond->priv = priv;
3015 	bond->is_bonded = is_bonded;
3016 	bond->port_map.port1 = v2p_p1;
3017 	bond->port_map.port2 = v2p_p2;
3018 	dev_hold(priv->dev);
3019 	queue_work(priv->mdev->workqueue, &bond->work);
3020 	return 0;
3021 }
3022 
3023 int mlx4_en_netdev_event(struct notifier_block *this,
3024 			 unsigned long event, void *ptr)
3025 {
3026 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3027 	u8 port = 0;
3028 	struct mlx4_en_dev *mdev;
3029 	struct mlx4_dev *dev;
3030 	int i, num_eth_ports = 0;
3031 	bool do_bond = true;
3032 	struct mlx4_en_priv *priv;
3033 	u8 v2p_port1 = 0;
3034 	u8 v2p_port2 = 0;
3035 
3036 	if (!net_eq(dev_net(ndev), &init_net))
3037 		return NOTIFY_DONE;
3038 
3039 	mdev = container_of(this, struct mlx4_en_dev, nb);
3040 	dev = mdev->dev;
3041 
3042 	/* Go into this mode only when two network devices set on two ports
3043 	 * of the same mlx4 device are slaves of the same bonding master
3044 	 */
3045 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
3046 		++num_eth_ports;
3047 		if (!port && (mdev->pndev[i] == ndev))
3048 			port = i;
3049 		mdev->upper[i] = mdev->pndev[i] ?
3050 			netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
3051 		/* condition not met: network device is a slave */
3052 		if (!mdev->upper[i])
3053 			do_bond = false;
3054 		if (num_eth_ports < 2)
3055 			continue;
3056 		/* condition not met: same master */
3057 		if (mdev->upper[i] != mdev->upper[i-1])
3058 			do_bond = false;
3059 	}
3060 	/* condition not met: 2 salves */
3061 	do_bond = (num_eth_ports ==  2) ? do_bond : false;
3062 
3063 	/* handle only events that come with enough info */
3064 	if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
3065 		return NOTIFY_DONE;
3066 
3067 	priv = netdev_priv(ndev);
3068 	if (do_bond) {
3069 		struct netdev_notifier_bonding_info *notifier_info = ptr;
3070 		struct netdev_bonding_info *bonding_info =
3071 			&notifier_info->bonding_info;
3072 
3073 		/* required mode 1, 2 or 4 */
3074 		if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3075 		    (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
3076 		    (bonding_info->master.bond_mode != BOND_MODE_8023AD))
3077 			do_bond = false;
3078 
3079 		/* require exactly 2 slaves */
3080 		if (bonding_info->master.num_slaves != 2)
3081 			do_bond = false;
3082 
3083 		/* calc v2p */
3084 		if (do_bond) {
3085 			if (bonding_info->master.bond_mode ==
3086 			    BOND_MODE_ACTIVEBACKUP) {
3087 				/* in active-backup mode virtual ports are
3088 				 * mapped to the physical port of the active
3089 				 * slave */
3090 				if (bonding_info->slave.state ==
3091 				    BOND_STATE_BACKUP) {
3092 					if (port == 1) {
3093 						v2p_port1 = 2;
3094 						v2p_port2 = 2;
3095 					} else {
3096 						v2p_port1 = 1;
3097 						v2p_port2 = 1;
3098 					}
3099 				} else { /* BOND_STATE_ACTIVE */
3100 					if (port == 1) {
3101 						v2p_port1 = 1;
3102 						v2p_port2 = 1;
3103 					} else {
3104 						v2p_port1 = 2;
3105 						v2p_port2 = 2;
3106 					}
3107 				}
3108 			} else { /* Active-Active */
3109 				/* in active-active mode a virtual port is
3110 				 * mapped to the native physical port if and only
3111 				 * if the physical port is up */
3112 				__s8 link = bonding_info->slave.link;
3113 
3114 				if (port == 1)
3115 					v2p_port2 = 2;
3116 				else
3117 					v2p_port1 = 1;
3118 				if ((link == BOND_LINK_UP) ||
3119 				    (link == BOND_LINK_FAIL)) {
3120 					if (port == 1)
3121 						v2p_port1 = 1;
3122 					else
3123 						v2p_port2 = 2;
3124 				} else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3125 					if (port == 1)
3126 						v2p_port1 = 2;
3127 					else
3128 						v2p_port2 = 1;
3129 				}
3130 			}
3131 		}
3132 	}
3133 
3134 	mlx4_en_queue_bond_work(priv, do_bond,
3135 				v2p_port1, v2p_port2);
3136 
3137 	return NOTIFY_DONE;
3138 }
3139 
3140 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3141 				     struct mlx4_en_stats_bitmap *stats_bitmap,
3142 				     u8 rx_ppp, u8 rx_pause,
3143 				     u8 tx_ppp, u8 tx_pause)
3144 {
3145 	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3146 
3147 	if (!mlx4_is_slave(dev) &&
3148 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3149 		mutex_lock(&stats_bitmap->mutex);
3150 		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3151 
3152 		if (rx_ppp)
3153 			bitmap_set(stats_bitmap->bitmap, last_i,
3154 				   NUM_FLOW_PRIORITY_STATS_RX);
3155 		last_i += NUM_FLOW_PRIORITY_STATS_RX;
3156 
3157 		if (rx_pause && !(rx_ppp))
3158 			bitmap_set(stats_bitmap->bitmap, last_i,
3159 				   NUM_FLOW_STATS_RX);
3160 		last_i += NUM_FLOW_STATS_RX;
3161 
3162 		if (tx_ppp)
3163 			bitmap_set(stats_bitmap->bitmap, last_i,
3164 				   NUM_FLOW_PRIORITY_STATS_TX);
3165 		last_i += NUM_FLOW_PRIORITY_STATS_TX;
3166 
3167 		if (tx_pause && !(tx_ppp))
3168 			bitmap_set(stats_bitmap->bitmap, last_i,
3169 				   NUM_FLOW_STATS_TX);
3170 		last_i += NUM_FLOW_STATS_TX;
3171 
3172 		mutex_unlock(&stats_bitmap->mutex);
3173 	}
3174 }
3175 
3176 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3177 			      struct mlx4_en_stats_bitmap *stats_bitmap,
3178 			      u8 rx_ppp, u8 rx_pause,
3179 			      u8 tx_ppp, u8 tx_pause)
3180 {
3181 	int last_i = 0;
3182 
3183 	mutex_init(&stats_bitmap->mutex);
3184 	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3185 
3186 	if (mlx4_is_slave(dev)) {
3187 		bitmap_set(stats_bitmap->bitmap, last_i +
3188 					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3189 		bitmap_set(stats_bitmap->bitmap, last_i +
3190 					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3191 		bitmap_set(stats_bitmap->bitmap, last_i +
3192 					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3193 		bitmap_set(stats_bitmap->bitmap, last_i +
3194 					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3195 		bitmap_set(stats_bitmap->bitmap, last_i +
3196 					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3197 		bitmap_set(stats_bitmap->bitmap, last_i +
3198 					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3199 	} else {
3200 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3201 	}
3202 	last_i += NUM_MAIN_STATS;
3203 
3204 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3205 	last_i += NUM_PORT_STATS;
3206 
3207 	if (mlx4_is_master(dev))
3208 		bitmap_set(stats_bitmap->bitmap, last_i,
3209 			   NUM_PF_STATS);
3210 	last_i += NUM_PF_STATS;
3211 
3212 	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3213 					rx_ppp, rx_pause,
3214 					tx_ppp, tx_pause);
3215 	last_i += NUM_FLOW_STATS;
3216 
3217 	if (!mlx4_is_slave(dev))
3218 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3219 	last_i += NUM_PKT_STATS;
3220 
3221 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3222 	last_i += NUM_XDP_STATS;
3223 
3224 	if (!mlx4_is_slave(dev))
3225 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PHY_STATS);
3226 	last_i += NUM_PHY_STATS;
3227 }
3228 
3229 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3230 			struct mlx4_en_port_profile *prof)
3231 {
3232 	struct net_device *dev;
3233 	struct mlx4_en_priv *priv;
3234 	int i, t;
3235 	int err;
3236 
3237 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3238 				 MAX_TX_RINGS, MAX_RX_RINGS);
3239 	if (dev == NULL)
3240 		return -ENOMEM;
3241 
3242 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3243 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3244 
3245 	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3246 	dev->dev_port = port - 1;
3247 
3248 	/*
3249 	 * Initialize driver private data
3250 	 */
3251 
3252 	priv = netdev_priv(dev);
3253 	memset(priv, 0, sizeof(struct mlx4_en_priv));
3254 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3255 	spin_lock_init(&priv->stats_lock);
3256 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3257 	INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
3258 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3259 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3260 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3261 	INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
3262 	INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
3263 #ifdef CONFIG_RFS_ACCEL
3264 	INIT_LIST_HEAD(&priv->filters);
3265 	spin_lock_init(&priv->filters_lock);
3266 #endif
3267 
3268 	priv->dev = dev;
3269 	priv->mdev = mdev;
3270 	priv->ddev = &mdev->pdev->dev;
3271 	priv->prof = prof;
3272 	priv->port = port;
3273 	priv->port_up = false;
3274 	priv->flags = prof->flags;
3275 	priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3276 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3277 			MLX4_WQE_CTRL_SOLICITED);
3278 	priv->num_tx_rings_p_up = mdev->profile.max_num_tx_rings_p_up;
3279 	priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3280 	netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3281 
3282 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3283 		priv->tx_ring_num[t] = prof->tx_ring_num[t];
3284 		if (!priv->tx_ring_num[t])
3285 			continue;
3286 
3287 		priv->tx_ring[t] = kcalloc(MAX_TX_RINGS,
3288 					   sizeof(struct mlx4_en_tx_ring *),
3289 					   GFP_KERNEL);
3290 		if (!priv->tx_ring[t]) {
3291 			err = -ENOMEM;
3292 			goto out;
3293 		}
3294 		priv->tx_cq[t] = kcalloc(MAX_TX_RINGS,
3295 					 sizeof(struct mlx4_en_cq *),
3296 					 GFP_KERNEL);
3297 		if (!priv->tx_cq[t]) {
3298 			err = -ENOMEM;
3299 			goto out;
3300 		}
3301 	}
3302 	priv->rx_ring_num = prof->rx_ring_num;
3303 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3304 	priv->cqe_size = mdev->dev->caps.cqe_size;
3305 	priv->mac_index = -1;
3306 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
3307 #ifdef CONFIG_MLX4_EN_DCB
3308 	if (!mlx4_is_slave(priv->mdev->dev)) {
3309 		u8 prio;
3310 
3311 		for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; ++prio) {
3312 			priv->ets.prio_tc[prio] = prio;
3313 			priv->ets.tc_tsa[prio]  = IEEE_8021QAZ_TSA_VENDOR;
3314 		}
3315 
3316 		priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3317 			DCB_CAP_DCBX_VER_IEEE;
3318 		priv->flags |= MLX4_EN_DCB_ENABLED;
3319 		priv->cee_config.pfc_state = false;
3320 
3321 		for (i = 0; i < MLX4_EN_NUM_UP_HIGH; i++)
3322 			priv->cee_config.dcb_pfc[i] = pfc_disabled;
3323 
3324 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3325 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3326 		} else {
3327 			en_info(priv, "enabling only PFC DCB ops\n");
3328 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3329 		}
3330 	}
3331 #endif
3332 
3333 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3334 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
3335 
3336 	/* Query for default mac and max mtu */
3337 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3338 
3339 	if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3340 	    MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3341 		priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3342 
3343 	/* Set default MAC */
3344 	dev->addr_len = ETH_ALEN;
3345 	mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3346 	if (!is_valid_ether_addr(dev->dev_addr)) {
3347 		en_err(priv, "Port: %d, invalid mac burned: %pM, quitting\n",
3348 		       priv->port, dev->dev_addr);
3349 		err = -EINVAL;
3350 		goto out;
3351 	} else if (mlx4_is_slave(priv->mdev->dev) &&
3352 		   (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3353 		/* Random MAC was assigned in mlx4_slave_cap
3354 		 * in mlx4_core module
3355 		 */
3356 		dev->addr_assign_type |= NET_ADDR_RANDOM;
3357 		en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3358 	}
3359 
3360 	memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3361 
3362 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3363 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3364 	err = mlx4_en_alloc_resources(priv);
3365 	if (err)
3366 		goto out;
3367 
3368 	/* Initialize time stamping config */
3369 	priv->hwtstamp_config.flags = 0;
3370 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3371 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3372 
3373 	/* Allocate page for receive rings */
3374 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3375 				MLX4_EN_PAGE_SIZE);
3376 	if (err) {
3377 		en_err(priv, "Failed to allocate page for rx qps\n");
3378 		goto out;
3379 	}
3380 	priv->allocated = 1;
3381 
3382 	/*
3383 	 * Initialize netdev entry points
3384 	 */
3385 	if (mlx4_is_master(priv->mdev->dev))
3386 		dev->netdev_ops = &mlx4_netdev_ops_master;
3387 	else
3388 		dev->netdev_ops = &mlx4_netdev_ops;
3389 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3390 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3391 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3392 
3393 	dev->ethtool_ops = &mlx4_en_ethtool_ops;
3394 
3395 	/*
3396 	 * Set driver features
3397 	 */
3398 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3399 	if (mdev->LSO_support)
3400 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3401 
3402 	if (mdev->dev->caps.tunnel_offload_mode ==
3403 	    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3404 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3405 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3406 				    NETIF_F_GSO_PARTIAL;
3407 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3408 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3409 				    NETIF_F_GSO_PARTIAL;
3410 		dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3411 		dev->hw_enc_features = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3412 				       NETIF_F_RXCSUM |
3413 				       NETIF_F_TSO | NETIF_F_TSO6 |
3414 				       NETIF_F_GSO_UDP_TUNNEL |
3415 				       NETIF_F_GSO_UDP_TUNNEL_CSUM |
3416 				       NETIF_F_GSO_PARTIAL;
3417 	}
3418 
3419 	dev->vlan_features = dev->hw_features;
3420 
3421 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3422 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3423 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3424 			NETIF_F_HW_VLAN_CTAG_FILTER;
3425 	dev->hw_features |= NETIF_F_LOOPBACK |
3426 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3427 
3428 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3429 		dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3430 			NETIF_F_HW_VLAN_STAG_FILTER;
3431 		dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3432 	}
3433 
3434 	if (mlx4_is_slave(mdev->dev)) {
3435 		bool vlan_offload_disabled;
3436 		int phv;
3437 
3438 		err = get_phv_bit(mdev->dev, port, &phv);
3439 		if (!err && phv) {
3440 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3441 			priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3442 		}
3443 		err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3444 							&vlan_offload_disabled);
3445 		if (!err && vlan_offload_disabled) {
3446 			dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3447 					      NETIF_F_HW_VLAN_CTAG_RX |
3448 					      NETIF_F_HW_VLAN_STAG_TX |
3449 					      NETIF_F_HW_VLAN_STAG_RX);
3450 			dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3451 					   NETIF_F_HW_VLAN_CTAG_RX |
3452 					   NETIF_F_HW_VLAN_STAG_TX |
3453 					   NETIF_F_HW_VLAN_STAG_RX);
3454 		}
3455 	} else {
3456 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3457 		    !(mdev->dev->caps.flags2 &
3458 		      MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3459 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3460 	}
3461 
3462 	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3463 		dev->hw_features |= NETIF_F_RXFCS;
3464 
3465 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3466 		dev->hw_features |= NETIF_F_RXALL;
3467 
3468 	if (mdev->dev->caps.steering_mode ==
3469 	    MLX4_STEERING_MODE_DEVICE_MANAGED &&
3470 	    mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3471 		dev->hw_features |= NETIF_F_NTUPLE;
3472 
3473 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3474 		dev->priv_flags |= IFF_UNICAST_FLT;
3475 
3476 	/* Setting a default hash function value */
3477 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3478 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3479 	} else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3480 		priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3481 	} else {
3482 		en_warn(priv,
3483 			"No RSS hash capabilities exposed, using Toeplitz\n");
3484 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3485 	}
3486 
3487 	/* MTU range: 68 - hw-specific max */
3488 	dev->min_mtu = ETH_MIN_MTU;
3489 	dev->max_mtu = priv->max_mtu;
3490 
3491 	mdev->pndev[port] = dev;
3492 	mdev->upper[port] = NULL;
3493 
3494 	netif_carrier_off(dev);
3495 	mlx4_en_set_default_moderation(priv);
3496 
3497 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3498 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3499 
3500 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3501 
3502 	/* Configure port */
3503 	mlx4_en_calc_rx_buf(dev);
3504 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3505 				    priv->rx_skb_size + ETH_FCS_LEN,
3506 				    prof->tx_pause, prof->tx_ppp,
3507 				    prof->rx_pause, prof->rx_ppp);
3508 	if (err) {
3509 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3510 		       priv->port, err);
3511 		goto out;
3512 	}
3513 
3514 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3515 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3516 		if (err) {
3517 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3518 			       err);
3519 			goto out;
3520 		}
3521 	}
3522 
3523 	/* Init port */
3524 	en_warn(priv, "Initializing port\n");
3525 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
3526 	if (err) {
3527 		en_err(priv, "Failed Initializing port\n");
3528 		goto out;
3529 	}
3530 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3531 
3532 	/* Initialize time stamp mechanism */
3533 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3534 		mlx4_en_init_timestamp(mdev);
3535 
3536 	queue_delayed_work(mdev->workqueue, &priv->service_task,
3537 			   SERVICE_TASK_DELAY);
3538 
3539 	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3540 				 mdev->profile.prof[priv->port].rx_ppp,
3541 				 mdev->profile.prof[priv->port].rx_pause,
3542 				 mdev->profile.prof[priv->port].tx_ppp,
3543 				 mdev->profile.prof[priv->port].tx_pause);
3544 
3545 	err = register_netdev(dev);
3546 	if (err) {
3547 		en_err(priv, "Netdev registration failed for port %d\n", port);
3548 		goto out;
3549 	}
3550 
3551 	priv->registered = 1;
3552 	devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3553 				  dev);
3554 
3555 	return 0;
3556 
3557 out:
3558 	mlx4_en_destroy_netdev(dev);
3559 	return err;
3560 }
3561 
3562 int mlx4_en_reset_config(struct net_device *dev,
3563 			 struct hwtstamp_config ts_config,
3564 			 netdev_features_t features)
3565 {
3566 	struct mlx4_en_priv *priv = netdev_priv(dev);
3567 	struct mlx4_en_dev *mdev = priv->mdev;
3568 	struct mlx4_en_port_profile new_prof;
3569 	struct mlx4_en_priv *tmp;
3570 	int port_up = 0;
3571 	int err = 0;
3572 
3573 	if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3574 	    priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3575 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3576 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3577 		return 0; /* Nothing to change */
3578 
3579 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3580 	    (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3581 	    (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3582 		en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3583 		return -EINVAL;
3584 	}
3585 
3586 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3587 	if (!tmp)
3588 		return -ENOMEM;
3589 
3590 	mutex_lock(&mdev->state_lock);
3591 
3592 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3593 	memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3594 
3595 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3596 	if (err)
3597 		goto out;
3598 
3599 	if (priv->port_up) {
3600 		port_up = 1;
3601 		mlx4_en_stop_port(dev, 1);
3602 	}
3603 
3604 	mlx4_en_safe_replace_resources(priv, tmp);
3605 
3606 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3607 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3608 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3609 		else
3610 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3611 	} else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3612 		/* RX time-stamping is OFF, update the RX vlan offload
3613 		 * to the latest wanted state
3614 		 */
3615 		if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3616 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3617 		else
3618 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3619 	}
3620 
3621 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3622 		if (features & NETIF_F_RXFCS)
3623 			dev->features |= NETIF_F_RXFCS;
3624 		else
3625 			dev->features &= ~NETIF_F_RXFCS;
3626 	}
3627 
3628 	/* RX vlan offload and RX time-stamping can't co-exist !
3629 	 * Regardless of the caller's choice,
3630 	 * Turn Off RX vlan offload in case of time-stamping is ON
3631 	 */
3632 	if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3633 		if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3634 			en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3635 		dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3636 	}
3637 
3638 	if (port_up) {
3639 		err = mlx4_en_start_port(dev);
3640 		if (err)
3641 			en_err(priv, "Failed starting port\n");
3642 	}
3643 
3644 out:
3645 	mutex_unlock(&mdev->state_lock);
3646 	kfree(tmp);
3647 	if (!err)
3648 		netdev_features_change(dev);
3649 	return err;
3650 }
3651