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 				SKB_DATA_ALIGN(sizeof(struct skb_shared_info))))
56 
57 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
58 {
59 	struct mlx4_en_priv *priv = netdev_priv(dev);
60 	int i;
61 	unsigned int offset = 0;
62 
63 	if (up && up != MLX4_EN_NUM_UP_HIGH)
64 		return -EINVAL;
65 
66 	netdev_set_num_tc(dev, up);
67 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
68 	/* Partition Tx queues evenly amongst UP's */
69 	for (i = 0; i < up; i++) {
70 		netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
71 		offset += priv->num_tx_rings_p_up;
72 	}
73 
74 #ifdef CONFIG_MLX4_EN_DCB
75 	if (!mlx4_is_slave(priv->mdev->dev)) {
76 		if (up) {
77 			if (priv->dcbx_cap)
78 				priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
79 		} else {
80 			priv->flags &= ~MLX4_EN_FLAG_DCB_ENABLED;
81 			priv->cee_config.pfc_state = false;
82 		}
83 	}
84 #endif /* CONFIG_MLX4_EN_DCB */
85 
86 	return 0;
87 }
88 
89 int mlx4_en_alloc_tx_queue_per_tc(struct net_device *dev, u8 tc)
90 {
91 	struct mlx4_en_priv *priv = netdev_priv(dev);
92 	struct mlx4_en_dev *mdev = priv->mdev;
93 	struct mlx4_en_port_profile new_prof;
94 	struct mlx4_en_priv *tmp;
95 	int total_count;
96 	int port_up = 0;
97 	int err = 0;
98 
99 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
100 	if (!tmp)
101 		return -ENOMEM;
102 
103 	mutex_lock(&mdev->state_lock);
104 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
105 	new_prof.num_up = (tc == 0) ? MLX4_EN_NUM_UP_LOW :
106 				      MLX4_EN_NUM_UP_HIGH;
107 	new_prof.tx_ring_num[TX] = new_prof.num_tx_rings_p_up *
108 				   new_prof.num_up;
109 	total_count = new_prof.tx_ring_num[TX] + new_prof.tx_ring_num[TX_XDP];
110 	if (total_count > MAX_TX_RINGS) {
111 		err = -EINVAL;
112 		en_err(priv,
113 		       "Total number of TX and XDP rings (%d) exceeds the maximum supported (%d)\n",
114 		       total_count, MAX_TX_RINGS);
115 		goto out;
116 	}
117 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
118 	if (err)
119 		goto out;
120 
121 	if (priv->port_up) {
122 		port_up = 1;
123 		mlx4_en_stop_port(dev, 1);
124 	}
125 
126 	mlx4_en_safe_replace_resources(priv, tmp);
127 	if (port_up) {
128 		err = mlx4_en_start_port(dev);
129 		if (err) {
130 			en_err(priv, "Failed starting port for setup TC\n");
131 			goto out;
132 		}
133 	}
134 
135 	err = mlx4_en_setup_tc(dev, tc);
136 out:
137 	mutex_unlock(&mdev->state_lock);
138 	kfree(tmp);
139 	return err;
140 }
141 
142 static int __mlx4_en_setup_tc(struct net_device *dev, enum tc_setup_type type,
143 			      void *type_data)
144 {
145 	struct tc_mqprio_qopt *mqprio = type_data;
146 
147 	if (type != TC_SETUP_QDISC_MQPRIO)
148 		return -EOPNOTSUPP;
149 
150 	if (mqprio->num_tc && mqprio->num_tc != MLX4_EN_NUM_UP_HIGH)
151 		return -EINVAL;
152 
153 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
154 
155 	return mlx4_en_alloc_tx_queue_per_tc(dev, mqprio->num_tc);
156 }
157 
158 #ifdef CONFIG_RFS_ACCEL
159 
160 struct mlx4_en_filter {
161 	struct list_head next;
162 	struct work_struct work;
163 
164 	u8     ip_proto;
165 	__be32 src_ip;
166 	__be32 dst_ip;
167 	__be16 src_port;
168 	__be16 dst_port;
169 
170 	int rxq_index;
171 	struct mlx4_en_priv *priv;
172 	u32 flow_id;			/* RFS infrastructure id */
173 	int id;				/* mlx4_en driver id */
174 	u64 reg_id;			/* Flow steering API id */
175 	u8 activated;			/* Used to prevent expiry before filter
176 					 * is attached
177 					 */
178 	struct hlist_node filter_chain;
179 };
180 
181 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
182 
183 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
184 {
185 	switch (ip_proto) {
186 	case IPPROTO_UDP:
187 		return MLX4_NET_TRANS_RULE_ID_UDP;
188 	case IPPROTO_TCP:
189 		return MLX4_NET_TRANS_RULE_ID_TCP;
190 	default:
191 		return MLX4_NET_TRANS_RULE_NUM;
192 	}
193 };
194 
195 /* Must not acquire state_lock, as its corresponding work_sync
196  * is done under it.
197  */
198 static void mlx4_en_filter_work(struct work_struct *work)
199 {
200 	struct mlx4_en_filter *filter = container_of(work,
201 						     struct mlx4_en_filter,
202 						     work);
203 	struct mlx4_en_priv *priv = filter->priv;
204 	struct mlx4_spec_list spec_tcp_udp = {
205 		.id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
206 		{
207 			.tcp_udp = {
208 				.dst_port = filter->dst_port,
209 				.dst_port_msk = (__force __be16)-1,
210 				.src_port = filter->src_port,
211 				.src_port_msk = (__force __be16)-1,
212 			},
213 		},
214 	};
215 	struct mlx4_spec_list spec_ip = {
216 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
217 		{
218 			.ipv4 = {
219 				.dst_ip = filter->dst_ip,
220 				.dst_ip_msk = (__force __be32)-1,
221 				.src_ip = filter->src_ip,
222 				.src_ip_msk = (__force __be32)-1,
223 			},
224 		},
225 	};
226 	struct mlx4_spec_list spec_eth = {
227 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
228 	};
229 	struct mlx4_net_trans_rule rule = {
230 		.list = LIST_HEAD_INIT(rule.list),
231 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
232 		.exclusive = 1,
233 		.allow_loopback = 1,
234 		.promisc_mode = MLX4_FS_REGULAR,
235 		.port = priv->port,
236 		.priority = MLX4_DOMAIN_RFS,
237 	};
238 	int rc;
239 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
240 
241 	if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
242 		en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
243 			filter->ip_proto);
244 		goto ignore;
245 	}
246 	list_add_tail(&spec_eth.list, &rule.list);
247 	list_add_tail(&spec_ip.list, &rule.list);
248 	list_add_tail(&spec_tcp_udp.list, &rule.list);
249 
250 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
251 	memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
252 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
253 
254 	filter->activated = 0;
255 
256 	if (filter->reg_id) {
257 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
258 		if (rc && rc != -ENOENT)
259 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
260 	}
261 
262 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
263 	if (rc)
264 		en_err(priv, "Error attaching flow. err = %d\n", rc);
265 
266 ignore:
267 	mlx4_en_filter_rfs_expire(priv);
268 
269 	filter->activated = 1;
270 }
271 
272 static inline struct hlist_head *
273 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
274 		   __be16 src_port, __be16 dst_port)
275 {
276 	unsigned long l;
277 	int bucket_idx;
278 
279 	l = (__force unsigned long)src_port |
280 	    ((__force unsigned long)dst_port << 2);
281 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
282 
283 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
284 
285 	return &priv->filter_hash[bucket_idx];
286 }
287 
288 static struct mlx4_en_filter *
289 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
290 		     __be32 dst_ip, u8 ip_proto, __be16 src_port,
291 		     __be16 dst_port, u32 flow_id)
292 {
293 	struct mlx4_en_filter *filter = NULL;
294 
295 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
296 	if (!filter)
297 		return NULL;
298 
299 	filter->priv = priv;
300 	filter->rxq_index = rxq_index;
301 	INIT_WORK(&filter->work, mlx4_en_filter_work);
302 
303 	filter->src_ip = src_ip;
304 	filter->dst_ip = dst_ip;
305 	filter->ip_proto = ip_proto;
306 	filter->src_port = src_port;
307 	filter->dst_port = dst_port;
308 
309 	filter->flow_id = flow_id;
310 
311 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
312 
313 	list_add_tail(&filter->next, &priv->filters);
314 	hlist_add_head(&filter->filter_chain,
315 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
316 					  dst_port));
317 
318 	return filter;
319 }
320 
321 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
322 {
323 	struct mlx4_en_priv *priv = filter->priv;
324 	int rc;
325 
326 	list_del(&filter->next);
327 
328 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
329 	if (rc && rc != -ENOENT)
330 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
331 
332 	kfree(filter);
333 }
334 
335 static inline struct mlx4_en_filter *
336 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
337 		    u8 ip_proto, __be16 src_port, __be16 dst_port)
338 {
339 	struct mlx4_en_filter *filter;
340 	struct mlx4_en_filter *ret = NULL;
341 
342 	hlist_for_each_entry(filter,
343 			     filter_hash_bucket(priv, src_ip, dst_ip,
344 						src_port, dst_port),
345 			     filter_chain) {
346 		if (filter->src_ip == src_ip &&
347 		    filter->dst_ip == dst_ip &&
348 		    filter->ip_proto == ip_proto &&
349 		    filter->src_port == src_port &&
350 		    filter->dst_port == dst_port) {
351 			ret = filter;
352 			break;
353 		}
354 	}
355 
356 	return ret;
357 }
358 
359 static int
360 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
361 		   u16 rxq_index, u32 flow_id)
362 {
363 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
364 	struct mlx4_en_filter *filter;
365 	const struct iphdr *ip;
366 	const __be16 *ports;
367 	u8 ip_proto;
368 	__be32 src_ip;
369 	__be32 dst_ip;
370 	__be16 src_port;
371 	__be16 dst_port;
372 	int nhoff = skb_network_offset(skb);
373 	int ret = 0;
374 
375 	if (skb->protocol != htons(ETH_P_IP))
376 		return -EPROTONOSUPPORT;
377 
378 	ip = (const struct iphdr *)(skb->data + nhoff);
379 	if (ip_is_fragment(ip))
380 		return -EPROTONOSUPPORT;
381 
382 	if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
383 		return -EPROTONOSUPPORT;
384 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
385 
386 	ip_proto = ip->protocol;
387 	src_ip = ip->saddr;
388 	dst_ip = ip->daddr;
389 	src_port = ports[0];
390 	dst_port = ports[1];
391 
392 	spin_lock_bh(&priv->filters_lock);
393 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
394 				     src_port, dst_port);
395 	if (filter) {
396 		if (filter->rxq_index == rxq_index)
397 			goto out;
398 
399 		filter->rxq_index = rxq_index;
400 	} else {
401 		filter = mlx4_en_filter_alloc(priv, rxq_index,
402 					      src_ip, dst_ip, ip_proto,
403 					      src_port, dst_port, flow_id);
404 		if (!filter) {
405 			ret = -ENOMEM;
406 			goto err;
407 		}
408 	}
409 
410 	queue_work(priv->mdev->workqueue, &filter->work);
411 
412 out:
413 	ret = filter->id;
414 err:
415 	spin_unlock_bh(&priv->filters_lock);
416 
417 	return ret;
418 }
419 
420 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
421 {
422 	struct mlx4_en_filter *filter, *tmp;
423 	LIST_HEAD(del_list);
424 
425 	spin_lock_bh(&priv->filters_lock);
426 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
427 		list_move(&filter->next, &del_list);
428 		hlist_del(&filter->filter_chain);
429 	}
430 	spin_unlock_bh(&priv->filters_lock);
431 
432 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
433 		cancel_work_sync(&filter->work);
434 		mlx4_en_filter_free(filter);
435 	}
436 }
437 
438 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
439 {
440 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
441 	LIST_HEAD(del_list);
442 	int i = 0;
443 
444 	spin_lock_bh(&priv->filters_lock);
445 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
446 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
447 			break;
448 
449 		if (filter->activated &&
450 		    !work_pending(&filter->work) &&
451 		    rps_may_expire_flow(priv->dev,
452 					filter->rxq_index, filter->flow_id,
453 					filter->id)) {
454 			list_move(&filter->next, &del_list);
455 			hlist_del(&filter->filter_chain);
456 		} else
457 			last_filter = filter;
458 
459 		i++;
460 	}
461 
462 	if (last_filter && (&last_filter->next != priv->filters.next))
463 		list_move(&priv->filters, &last_filter->next);
464 
465 	spin_unlock_bh(&priv->filters_lock);
466 
467 	list_for_each_entry_safe(filter, tmp, &del_list, next)
468 		mlx4_en_filter_free(filter);
469 }
470 #endif
471 
472 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
473 				   __be16 proto, u16 vid)
474 {
475 	struct mlx4_en_priv *priv = netdev_priv(dev);
476 	struct mlx4_en_dev *mdev = priv->mdev;
477 	int err;
478 	int idx;
479 
480 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
481 
482 	set_bit(vid, priv->active_vlans);
483 
484 	/* Add VID to port VLAN filter */
485 	mutex_lock(&mdev->state_lock);
486 	if (mdev->device_up && priv->port_up) {
487 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
488 		if (err) {
489 			en_err(priv, "Failed configuring VLAN filter\n");
490 			goto out;
491 		}
492 	}
493 	err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
494 	if (err)
495 		en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
496 
497 out:
498 	mutex_unlock(&mdev->state_lock);
499 	return err;
500 }
501 
502 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
503 				    __be16 proto, u16 vid)
504 {
505 	struct mlx4_en_priv *priv = netdev_priv(dev);
506 	struct mlx4_en_dev *mdev = priv->mdev;
507 	int err = 0;
508 
509 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
510 
511 	clear_bit(vid, priv->active_vlans);
512 
513 	/* Remove VID from port VLAN filter */
514 	mutex_lock(&mdev->state_lock);
515 	mlx4_unregister_vlan(mdev->dev, priv->port, vid);
516 
517 	if (mdev->device_up && priv->port_up) {
518 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
519 		if (err)
520 			en_err(priv, "Failed configuring VLAN filter\n");
521 	}
522 	mutex_unlock(&mdev->state_lock);
523 
524 	return err;
525 }
526 
527 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
528 {
529 	int i;
530 	for (i = ETH_ALEN - 1; i >= 0; --i) {
531 		dst_mac[i] = src_mac & 0xff;
532 		src_mac >>= 8;
533 	}
534 	memset(&dst_mac[ETH_ALEN], 0, 2);
535 }
536 
537 
538 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
539 				    int qpn, u64 *reg_id)
540 {
541 	int err;
542 
543 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
544 	    priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
545 		return 0; /* do nothing */
546 
547 	err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
548 				    MLX4_DOMAIN_NIC, reg_id);
549 	if (err) {
550 		en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
551 		return err;
552 	}
553 	en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
554 	return 0;
555 }
556 
557 
558 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
559 				unsigned char *mac, int *qpn, u64 *reg_id)
560 {
561 	struct mlx4_en_dev *mdev = priv->mdev;
562 	struct mlx4_dev *dev = mdev->dev;
563 	int err;
564 
565 	switch (dev->caps.steering_mode) {
566 	case MLX4_STEERING_MODE_B0: {
567 		struct mlx4_qp qp;
568 		u8 gid[16] = {0};
569 
570 		qp.qpn = *qpn;
571 		memcpy(&gid[10], mac, ETH_ALEN);
572 		gid[5] = priv->port;
573 
574 		err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
575 		break;
576 	}
577 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
578 		struct mlx4_spec_list spec_eth = { {NULL} };
579 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
580 
581 		struct mlx4_net_trans_rule rule = {
582 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
583 			.exclusive = 0,
584 			.allow_loopback = 1,
585 			.promisc_mode = MLX4_FS_REGULAR,
586 			.priority = MLX4_DOMAIN_NIC,
587 		};
588 
589 		rule.port = priv->port;
590 		rule.qpn = *qpn;
591 		INIT_LIST_HEAD(&rule.list);
592 
593 		spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
594 		memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
595 		memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
596 		list_add_tail(&spec_eth.list, &rule.list);
597 
598 		err = mlx4_flow_attach(dev, &rule, reg_id);
599 		break;
600 	}
601 	default:
602 		return -EINVAL;
603 	}
604 	if (err)
605 		en_warn(priv, "Failed Attaching Unicast\n");
606 
607 	return err;
608 }
609 
610 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
611 				     unsigned char *mac, int qpn, u64 reg_id)
612 {
613 	struct mlx4_en_dev *mdev = priv->mdev;
614 	struct mlx4_dev *dev = mdev->dev;
615 
616 	switch (dev->caps.steering_mode) {
617 	case MLX4_STEERING_MODE_B0: {
618 		struct mlx4_qp qp;
619 		u8 gid[16] = {0};
620 
621 		qp.qpn = qpn;
622 		memcpy(&gid[10], mac, ETH_ALEN);
623 		gid[5] = priv->port;
624 
625 		mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
626 		break;
627 	}
628 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
629 		mlx4_flow_detach(dev, reg_id);
630 		break;
631 	}
632 	default:
633 		en_err(priv, "Invalid steering mode.\n");
634 	}
635 }
636 
637 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
638 {
639 	struct mlx4_en_dev *mdev = priv->mdev;
640 	struct mlx4_dev *dev = mdev->dev;
641 	int index = 0;
642 	int err = 0;
643 	int *qpn = &priv->base_qpn;
644 	u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
645 
646 	en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
647 	       priv->dev->dev_addr);
648 	index = mlx4_register_mac(dev, priv->port, mac);
649 	if (index < 0) {
650 		err = index;
651 		en_err(priv, "Failed adding MAC: %pM\n",
652 		       priv->dev->dev_addr);
653 		return err;
654 	}
655 
656 	en_info(priv, "Steering Mode %d\n", dev->caps.steering_mode);
657 
658 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
659 		int base_qpn = mlx4_get_base_qpn(dev, priv->port);
660 		*qpn = base_qpn + index;
661 		return 0;
662 	}
663 
664 	err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP,
665 				    MLX4_RES_USAGE_DRIVER);
666 	en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
667 	if (err) {
668 		en_err(priv, "Failed to reserve qp for mac registration\n");
669 		mlx4_unregister_mac(dev, priv->port, mac);
670 		return err;
671 	}
672 
673 	return 0;
674 }
675 
676 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
677 {
678 	struct mlx4_en_dev *mdev = priv->mdev;
679 	struct mlx4_dev *dev = mdev->dev;
680 	int qpn = priv->base_qpn;
681 
682 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
683 		u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
684 		en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
685 		       priv->dev->dev_addr);
686 		mlx4_unregister_mac(dev, priv->port, mac);
687 	} else {
688 		en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
689 		       priv->port, qpn);
690 		mlx4_qp_release_range(dev, qpn, 1);
691 		priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
692 	}
693 }
694 
695 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
696 			       unsigned char *new_mac, unsigned char *prev_mac)
697 {
698 	struct mlx4_en_dev *mdev = priv->mdev;
699 	struct mlx4_dev *dev = mdev->dev;
700 	int err = 0;
701 	u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
702 
703 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
704 		struct hlist_head *bucket;
705 		unsigned int mac_hash;
706 		struct mlx4_mac_entry *entry;
707 		struct hlist_node *tmp;
708 		u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
709 
710 		bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
711 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
712 			if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
713 				mlx4_en_uc_steer_release(priv, entry->mac,
714 							 qpn, entry->reg_id);
715 				mlx4_unregister_mac(dev, priv->port,
716 						    prev_mac_u64);
717 				hlist_del_rcu(&entry->hlist);
718 				synchronize_rcu();
719 				memcpy(entry->mac, new_mac, ETH_ALEN);
720 				entry->reg_id = 0;
721 				mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
722 				hlist_add_head_rcu(&entry->hlist,
723 						   &priv->mac_hash[mac_hash]);
724 				mlx4_register_mac(dev, priv->port, new_mac_u64);
725 				err = mlx4_en_uc_steer_add(priv, new_mac,
726 							   &qpn,
727 							   &entry->reg_id);
728 				if (err)
729 					return err;
730 				if (priv->tunnel_reg_id) {
731 					mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
732 					priv->tunnel_reg_id = 0;
733 				}
734 				err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
735 							       &priv->tunnel_reg_id);
736 				return err;
737 			}
738 		}
739 		return -EINVAL;
740 	}
741 
742 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
743 }
744 
745 static void mlx4_en_update_user_mac(struct mlx4_en_priv *priv,
746 				    unsigned char new_mac[ETH_ALEN + 2])
747 {
748 	struct mlx4_en_dev *mdev = priv->mdev;
749 	int err;
750 
751 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_USER_MAC_EN))
752 		return;
753 
754 	err = mlx4_SET_PORT_user_mac(mdev->dev, priv->port, new_mac);
755 	if (err)
756 		en_err(priv, "Failed to pass user MAC(%pM) to Firmware for port %d, with error %d\n",
757 		       new_mac, priv->port, err);
758 }
759 
760 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
761 			      unsigned char new_mac[ETH_ALEN + 2])
762 {
763 	int err = 0;
764 
765 	if (priv->port_up) {
766 		/* Remove old MAC and insert the new one */
767 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
768 					  new_mac, priv->current_mac);
769 		if (err)
770 			en_err(priv, "Failed changing HW MAC address\n");
771 	} else
772 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
773 
774 	if (!err)
775 		memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
776 
777 	return err;
778 }
779 
780 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
781 {
782 	struct mlx4_en_priv *priv = netdev_priv(dev);
783 	struct mlx4_en_dev *mdev = priv->mdev;
784 	struct sockaddr *saddr = addr;
785 	unsigned char new_mac[ETH_ALEN + 2];
786 	int err;
787 
788 	if (!is_valid_ether_addr(saddr->sa_data))
789 		return -EADDRNOTAVAIL;
790 
791 	mutex_lock(&mdev->state_lock);
792 	memcpy(new_mac, saddr->sa_data, ETH_ALEN);
793 	err = mlx4_en_do_set_mac(priv, new_mac);
794 	if (err)
795 		goto out;
796 
797 	memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
798 	mlx4_en_update_user_mac(priv, new_mac);
799 out:
800 	mutex_unlock(&mdev->state_lock);
801 
802 	return err;
803 }
804 
805 static void mlx4_en_clear_list(struct net_device *dev)
806 {
807 	struct mlx4_en_priv *priv = netdev_priv(dev);
808 	struct mlx4_en_mc_list *tmp, *mc_to_del;
809 
810 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
811 		list_del(&mc_to_del->list);
812 		kfree(mc_to_del);
813 	}
814 }
815 
816 static void mlx4_en_cache_mclist(struct net_device *dev)
817 {
818 	struct mlx4_en_priv *priv = netdev_priv(dev);
819 	struct netdev_hw_addr *ha;
820 	struct mlx4_en_mc_list *tmp;
821 
822 	mlx4_en_clear_list(dev);
823 	netdev_for_each_mc_addr(ha, dev) {
824 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
825 		if (!tmp) {
826 			mlx4_en_clear_list(dev);
827 			return;
828 		}
829 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
830 		list_add_tail(&tmp->list, &priv->mc_list);
831 	}
832 }
833 
834 static void update_mclist_flags(struct mlx4_en_priv *priv,
835 				struct list_head *dst,
836 				struct list_head *src)
837 {
838 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
839 	bool found;
840 
841 	/* Find all the entries that should be removed from dst,
842 	 * These are the entries that are not found in src
843 	 */
844 	list_for_each_entry(dst_tmp, dst, list) {
845 		found = false;
846 		list_for_each_entry(src_tmp, src, list) {
847 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
848 				found = true;
849 				break;
850 			}
851 		}
852 		if (!found)
853 			dst_tmp->action = MCLIST_REM;
854 	}
855 
856 	/* Add entries that exist in src but not in dst
857 	 * mark them as need to add
858 	 */
859 	list_for_each_entry(src_tmp, src, list) {
860 		found = false;
861 		list_for_each_entry(dst_tmp, dst, list) {
862 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
863 				dst_tmp->action = MCLIST_NONE;
864 				found = true;
865 				break;
866 			}
867 		}
868 		if (!found) {
869 			new_mc = kmemdup(src_tmp,
870 					 sizeof(struct mlx4_en_mc_list),
871 					 GFP_KERNEL);
872 			if (!new_mc)
873 				return;
874 
875 			new_mc->action = MCLIST_ADD;
876 			list_add_tail(&new_mc->list, dst);
877 		}
878 	}
879 }
880 
881 static void mlx4_en_set_rx_mode(struct net_device *dev)
882 {
883 	struct mlx4_en_priv *priv = netdev_priv(dev);
884 
885 	if (!priv->port_up)
886 		return;
887 
888 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
889 }
890 
891 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
892 				     struct mlx4_en_dev *mdev)
893 {
894 	int err = 0;
895 
896 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
897 		if (netif_msg_rx_status(priv))
898 			en_warn(priv, "Entering promiscuous mode\n");
899 		priv->flags |= MLX4_EN_FLAG_PROMISC;
900 
901 		/* Enable promiscouos mode */
902 		switch (mdev->dev->caps.steering_mode) {
903 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
904 			err = mlx4_flow_steer_promisc_add(mdev->dev,
905 							  priv->port,
906 							  priv->base_qpn,
907 							  MLX4_FS_ALL_DEFAULT);
908 			if (err)
909 				en_err(priv, "Failed enabling promiscuous mode\n");
910 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
911 			break;
912 
913 		case MLX4_STEERING_MODE_B0:
914 			err = mlx4_unicast_promisc_add(mdev->dev,
915 						       priv->base_qpn,
916 						       priv->port);
917 			if (err)
918 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
919 
920 			/* Add the default qp number as multicast
921 			 * promisc
922 			 */
923 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
924 				err = mlx4_multicast_promisc_add(mdev->dev,
925 								 priv->base_qpn,
926 								 priv->port);
927 				if (err)
928 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
929 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
930 			}
931 			break;
932 
933 		case MLX4_STEERING_MODE_A0:
934 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
935 						     priv->port,
936 						     priv->base_qpn,
937 						     1);
938 			if (err)
939 				en_err(priv, "Failed enabling promiscuous mode\n");
940 			break;
941 		}
942 
943 		/* Disable port multicast filter (unconditionally) */
944 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
945 					  0, MLX4_MCAST_DISABLE);
946 		if (err)
947 			en_err(priv, "Failed disabling multicast filter\n");
948 	}
949 }
950 
951 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
952 				       struct mlx4_en_dev *mdev)
953 {
954 	int err = 0;
955 
956 	if (netif_msg_rx_status(priv))
957 		en_warn(priv, "Leaving promiscuous mode\n");
958 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
959 
960 	/* Disable promiscouos mode */
961 	switch (mdev->dev->caps.steering_mode) {
962 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
963 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
964 						     priv->port,
965 						     MLX4_FS_ALL_DEFAULT);
966 		if (err)
967 			en_err(priv, "Failed disabling promiscuous mode\n");
968 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
969 		break;
970 
971 	case MLX4_STEERING_MODE_B0:
972 		err = mlx4_unicast_promisc_remove(mdev->dev,
973 						  priv->base_qpn,
974 						  priv->port);
975 		if (err)
976 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
977 		/* Disable Multicast promisc */
978 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
979 			err = mlx4_multicast_promisc_remove(mdev->dev,
980 							    priv->base_qpn,
981 							    priv->port);
982 			if (err)
983 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
984 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
985 		}
986 		break;
987 
988 	case MLX4_STEERING_MODE_A0:
989 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
990 					     priv->port,
991 					     priv->base_qpn, 0);
992 		if (err)
993 			en_err(priv, "Failed disabling promiscuous mode\n");
994 		break;
995 	}
996 }
997 
998 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
999 				 struct net_device *dev,
1000 				 struct mlx4_en_dev *mdev)
1001 {
1002 	struct mlx4_en_mc_list *mclist, *tmp;
1003 	u64 mcast_addr = 0;
1004 	u8 mc_list[16] = {0};
1005 	int err = 0;
1006 
1007 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
1008 	if (dev->flags & IFF_ALLMULTI) {
1009 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1010 					  0, MLX4_MCAST_DISABLE);
1011 		if (err)
1012 			en_err(priv, "Failed disabling multicast filter\n");
1013 
1014 		/* Add the default qp number as multicast promisc */
1015 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1016 			switch (mdev->dev->caps.steering_mode) {
1017 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1018 				err = mlx4_flow_steer_promisc_add(mdev->dev,
1019 								  priv->port,
1020 								  priv->base_qpn,
1021 								  MLX4_FS_MC_DEFAULT);
1022 				break;
1023 
1024 			case MLX4_STEERING_MODE_B0:
1025 				err = mlx4_multicast_promisc_add(mdev->dev,
1026 								 priv->base_qpn,
1027 								 priv->port);
1028 				break;
1029 
1030 			case MLX4_STEERING_MODE_A0:
1031 				break;
1032 			}
1033 			if (err)
1034 				en_err(priv, "Failed entering multicast promisc mode\n");
1035 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1036 		}
1037 	} else {
1038 		/* Disable Multicast promisc */
1039 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1040 			switch (mdev->dev->caps.steering_mode) {
1041 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1042 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
1043 								     priv->port,
1044 								     MLX4_FS_MC_DEFAULT);
1045 				break;
1046 
1047 			case MLX4_STEERING_MODE_B0:
1048 				err = mlx4_multicast_promisc_remove(mdev->dev,
1049 								    priv->base_qpn,
1050 								    priv->port);
1051 				break;
1052 
1053 			case MLX4_STEERING_MODE_A0:
1054 				break;
1055 			}
1056 			if (err)
1057 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
1058 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1059 		}
1060 
1061 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1062 					  0, MLX4_MCAST_DISABLE);
1063 		if (err)
1064 			en_err(priv, "Failed disabling multicast filter\n");
1065 
1066 		/* Flush mcast filter and init it with broadcast address */
1067 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1068 				    1, MLX4_MCAST_CONFIG);
1069 
1070 		/* Update multicast list - we cache all addresses so they won't
1071 		 * change while HW is updated holding the command semaphor */
1072 		netif_addr_lock_bh(dev);
1073 		mlx4_en_cache_mclist(dev);
1074 		netif_addr_unlock_bh(dev);
1075 		list_for_each_entry(mclist, &priv->mc_list, list) {
1076 			mcast_addr = mlx4_mac_to_u64(mclist->addr);
1077 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1078 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
1079 		}
1080 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1081 					  0, MLX4_MCAST_ENABLE);
1082 		if (err)
1083 			en_err(priv, "Failed enabling multicast filter\n");
1084 
1085 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1086 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1087 			if (mclist->action == MCLIST_REM) {
1088 				/* detach this address and delete from list */
1089 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1090 				mc_list[5] = priv->port;
1091 				err = mlx4_multicast_detach(mdev->dev,
1092 							    priv->rss_map.indir_qp,
1093 							    mc_list,
1094 							    MLX4_PROT_ETH,
1095 							    mclist->reg_id);
1096 				if (err)
1097 					en_err(priv, "Fail to detach multicast address\n");
1098 
1099 				if (mclist->tunnel_reg_id) {
1100 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1101 					if (err)
1102 						en_err(priv, "Failed to detach multicast address\n");
1103 				}
1104 
1105 				/* remove from list */
1106 				list_del(&mclist->list);
1107 				kfree(mclist);
1108 			} else if (mclist->action == MCLIST_ADD) {
1109 				/* attach the address */
1110 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1111 				/* needed for B0 steering support */
1112 				mc_list[5] = priv->port;
1113 				err = mlx4_multicast_attach(mdev->dev,
1114 							    priv->rss_map.indir_qp,
1115 							    mc_list,
1116 							    priv->port, 0,
1117 							    MLX4_PROT_ETH,
1118 							    &mclist->reg_id);
1119 				if (err)
1120 					en_err(priv, "Fail to attach multicast address\n");
1121 
1122 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1123 							       &mclist->tunnel_reg_id);
1124 				if (err)
1125 					en_err(priv, "Failed to attach multicast address\n");
1126 			}
1127 		}
1128 	}
1129 }
1130 
1131 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1132 				 struct net_device *dev,
1133 				 struct mlx4_en_dev *mdev)
1134 {
1135 	struct netdev_hw_addr *ha;
1136 	struct mlx4_mac_entry *entry;
1137 	struct hlist_node *tmp;
1138 	bool found;
1139 	u64 mac;
1140 	int err = 0;
1141 	struct hlist_head *bucket;
1142 	unsigned int i;
1143 	int removed = 0;
1144 	u32 prev_flags;
1145 
1146 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1147 	 * since all modification code is protected by mdev->state_lock
1148 	 */
1149 
1150 	/* find what to remove */
1151 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1152 		bucket = &priv->mac_hash[i];
1153 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1154 			found = false;
1155 			netdev_for_each_uc_addr(ha, dev) {
1156 				if (ether_addr_equal_64bits(entry->mac,
1157 							    ha->addr)) {
1158 					found = true;
1159 					break;
1160 				}
1161 			}
1162 
1163 			/* MAC address of the port is not in uc list */
1164 			if (ether_addr_equal_64bits(entry->mac,
1165 						    priv->current_mac))
1166 				found = true;
1167 
1168 			if (!found) {
1169 				mac = mlx4_mac_to_u64(entry->mac);
1170 				mlx4_en_uc_steer_release(priv, entry->mac,
1171 							 priv->base_qpn,
1172 							 entry->reg_id);
1173 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1174 
1175 				hlist_del_rcu(&entry->hlist);
1176 				kfree_rcu(entry, rcu);
1177 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1178 				       entry->mac, priv->port);
1179 				++removed;
1180 			}
1181 		}
1182 	}
1183 
1184 	/* if we didn't remove anything, there is no use in trying to add
1185 	 * again once we are in a forced promisc mode state
1186 	 */
1187 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1188 		return;
1189 
1190 	prev_flags = priv->flags;
1191 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1192 
1193 	/* find what to add */
1194 	netdev_for_each_uc_addr(ha, dev) {
1195 		found = false;
1196 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1197 		hlist_for_each_entry(entry, bucket, hlist) {
1198 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1199 				found = true;
1200 				break;
1201 			}
1202 		}
1203 
1204 		if (!found) {
1205 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1206 			if (!entry) {
1207 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1208 				       ha->addr, priv->port);
1209 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1210 				break;
1211 			}
1212 			mac = mlx4_mac_to_u64(ha->addr);
1213 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1214 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1215 			if (err < 0) {
1216 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1217 				       ha->addr, priv->port, err);
1218 				kfree(entry);
1219 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1220 				break;
1221 			}
1222 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1223 						   &priv->base_qpn,
1224 						   &entry->reg_id);
1225 			if (err) {
1226 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1227 				       ha->addr, priv->port, err);
1228 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1229 				kfree(entry);
1230 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1231 				break;
1232 			} else {
1233 				unsigned int mac_hash;
1234 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1235 				       ha->addr, priv->port);
1236 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1237 				bucket = &priv->mac_hash[mac_hash];
1238 				hlist_add_head_rcu(&entry->hlist, bucket);
1239 			}
1240 		}
1241 	}
1242 
1243 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1244 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1245 			priv->port);
1246 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1247 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1248 			priv->port);
1249 	}
1250 }
1251 
1252 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1253 {
1254 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1255 						 rx_mode_task);
1256 	struct mlx4_en_dev *mdev = priv->mdev;
1257 	struct net_device *dev = priv->dev;
1258 
1259 	mutex_lock(&mdev->state_lock);
1260 	if (!mdev->device_up) {
1261 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1262 		goto out;
1263 	}
1264 	if (!priv->port_up) {
1265 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1266 		goto out;
1267 	}
1268 
1269 	if (!netif_carrier_ok(dev)) {
1270 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1271 			if (priv->port_state.link_state) {
1272 				priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1273 				netif_carrier_on(dev);
1274 				en_dbg(LINK, priv, "Link Up\n");
1275 			}
1276 		}
1277 	}
1278 
1279 	if (dev->priv_flags & IFF_UNICAST_FLT)
1280 		mlx4_en_do_uc_filter(priv, dev, mdev);
1281 
1282 	/* Promsicuous mode: disable all filters */
1283 	if ((dev->flags & IFF_PROMISC) ||
1284 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1285 		mlx4_en_set_promisc_mode(priv, mdev);
1286 		goto out;
1287 	}
1288 
1289 	/* Not in promiscuous mode */
1290 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1291 		mlx4_en_clear_promisc_mode(priv, mdev);
1292 
1293 	mlx4_en_do_multicast(priv, dev, mdev);
1294 out:
1295 	mutex_unlock(&mdev->state_lock);
1296 }
1297 
1298 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1299 {
1300 	u64 reg_id;
1301 	int err = 0;
1302 	int *qpn = &priv->base_qpn;
1303 	struct mlx4_mac_entry *entry;
1304 
1305 	err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1306 	if (err)
1307 		return err;
1308 
1309 	err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1310 				       &priv->tunnel_reg_id);
1311 	if (err)
1312 		goto tunnel_err;
1313 
1314 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1315 	if (!entry) {
1316 		err = -ENOMEM;
1317 		goto alloc_err;
1318 	}
1319 
1320 	memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1321 	memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1322 	entry->reg_id = reg_id;
1323 	hlist_add_head_rcu(&entry->hlist,
1324 			   &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1325 
1326 	return 0;
1327 
1328 alloc_err:
1329 	if (priv->tunnel_reg_id)
1330 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1331 
1332 tunnel_err:
1333 	mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1334 	return err;
1335 }
1336 
1337 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1338 {
1339 	u64 mac;
1340 	unsigned int i;
1341 	int qpn = priv->base_qpn;
1342 	struct hlist_head *bucket;
1343 	struct hlist_node *tmp;
1344 	struct mlx4_mac_entry *entry;
1345 
1346 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1347 		bucket = &priv->mac_hash[i];
1348 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1349 			mac = mlx4_mac_to_u64(entry->mac);
1350 			en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1351 			       entry->mac);
1352 			mlx4_en_uc_steer_release(priv, entry->mac,
1353 						 qpn, entry->reg_id);
1354 
1355 			mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1356 			hlist_del_rcu(&entry->hlist);
1357 			kfree_rcu(entry, rcu);
1358 		}
1359 	}
1360 
1361 	if (priv->tunnel_reg_id) {
1362 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1363 		priv->tunnel_reg_id = 0;
1364 	}
1365 }
1366 
1367 static void mlx4_en_tx_timeout(struct net_device *dev, unsigned int txqueue)
1368 {
1369 	struct mlx4_en_priv *priv = netdev_priv(dev);
1370 	struct mlx4_en_dev *mdev = priv->mdev;
1371 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][txqueue];
1372 
1373 	if (netif_msg_timer(priv))
1374 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1375 
1376 	en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1377 		txqueue, tx_ring->qpn, tx_ring->sp_cqn,
1378 		tx_ring->cons, tx_ring->prod);
1379 
1380 	priv->port_stats.tx_timeout++;
1381 	if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state)) {
1382 		en_dbg(DRV, priv, "Scheduling port restart\n");
1383 		queue_work(mdev->workqueue, &priv->restart_task);
1384 	}
1385 }
1386 
1387 
1388 static void
1389 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1390 {
1391 	struct mlx4_en_priv *priv = netdev_priv(dev);
1392 
1393 	spin_lock_bh(&priv->stats_lock);
1394 	mlx4_en_fold_software_stats(dev);
1395 	netdev_stats_to_stats64(stats, &dev->stats);
1396 	spin_unlock_bh(&priv->stats_lock);
1397 }
1398 
1399 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1400 {
1401 	struct mlx4_en_cq *cq;
1402 	int i, t;
1403 
1404 	/* If we haven't received a specific coalescing setting
1405 	 * (module param), we set the moderation parameters as follows:
1406 	 * - moder_cnt is set to the number of mtu sized packets to
1407 	 *   satisfy our coalescing target.
1408 	 * - moder_time is set to a fixed value.
1409 	 */
1410 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1411 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1412 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1413 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1414 	en_dbg(INTR, priv, "Default coalescing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1415 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1416 
1417 	/* Setup cq moderation params */
1418 	for (i = 0; i < priv->rx_ring_num; i++) {
1419 		cq = priv->rx_cq[i];
1420 		cq->moder_cnt = priv->rx_frames;
1421 		cq->moder_time = priv->rx_usecs;
1422 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1423 		priv->last_moder_packets[i] = 0;
1424 		priv->last_moder_bytes[i] = 0;
1425 	}
1426 
1427 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1428 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1429 			cq = priv->tx_cq[t][i];
1430 			cq->moder_cnt = priv->tx_frames;
1431 			cq->moder_time = priv->tx_usecs;
1432 		}
1433 	}
1434 
1435 	/* Reset auto-moderation params */
1436 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1437 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1438 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1439 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1440 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1441 	priv->adaptive_rx_coal = 1;
1442 	priv->last_moder_jiffies = 0;
1443 	priv->last_moder_tx_packets = 0;
1444 }
1445 
1446 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1447 {
1448 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1449 	u32 pkt_rate_high, pkt_rate_low;
1450 	struct mlx4_en_cq *cq;
1451 	unsigned long packets;
1452 	unsigned long rate;
1453 	unsigned long avg_pkt_size;
1454 	unsigned long rx_packets;
1455 	unsigned long rx_bytes;
1456 	unsigned long rx_pkt_diff;
1457 	int moder_time;
1458 	int ring, err;
1459 
1460 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1461 		return;
1462 
1463 	pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1464 	pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1465 
1466 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1467 		rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1468 		rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1469 
1470 		rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1471 		packets = rx_pkt_diff;
1472 		rate = packets * HZ / period;
1473 		avg_pkt_size = packets ? (rx_bytes -
1474 				priv->last_moder_bytes[ring]) / packets : 0;
1475 
1476 		/* Apply auto-moderation only when packet rate
1477 		 * exceeds a rate that it matters */
1478 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1479 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1480 			if (rate <= pkt_rate_low)
1481 				moder_time = priv->rx_usecs_low;
1482 			else if (rate >= pkt_rate_high)
1483 				moder_time = priv->rx_usecs_high;
1484 			else
1485 				moder_time = (rate - pkt_rate_low) *
1486 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1487 					(pkt_rate_high - pkt_rate_low) +
1488 					priv->rx_usecs_low;
1489 		} else {
1490 			moder_time = priv->rx_usecs_low;
1491 		}
1492 
1493 		cq = priv->rx_cq[ring];
1494 		if (moder_time != priv->last_moder_time[ring] ||
1495 		    cq->moder_cnt != priv->rx_frames) {
1496 			priv->last_moder_time[ring] = moder_time;
1497 			cq->moder_time = moder_time;
1498 			cq->moder_cnt = priv->rx_frames;
1499 			err = mlx4_en_set_cq_moder(priv, cq);
1500 			if (err)
1501 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1502 				       ring);
1503 		}
1504 		priv->last_moder_packets[ring] = rx_packets;
1505 		priv->last_moder_bytes[ring] = rx_bytes;
1506 	}
1507 
1508 	priv->last_moder_jiffies = jiffies;
1509 }
1510 
1511 static void mlx4_en_do_get_stats(struct work_struct *work)
1512 {
1513 	struct delayed_work *delay = to_delayed_work(work);
1514 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1515 						 stats_task);
1516 	struct mlx4_en_dev *mdev = priv->mdev;
1517 	int err;
1518 
1519 	mutex_lock(&mdev->state_lock);
1520 	if (mdev->device_up) {
1521 		if (priv->port_up) {
1522 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1523 			if (err)
1524 				en_dbg(HW, priv, "Could not update stats\n");
1525 
1526 			mlx4_en_auto_moderation(priv);
1527 		}
1528 
1529 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1530 	}
1531 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1532 		mlx4_en_do_set_mac(priv, priv->current_mac);
1533 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1534 	}
1535 	mutex_unlock(&mdev->state_lock);
1536 }
1537 
1538 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1539  * periodically
1540  */
1541 static void mlx4_en_service_task(struct work_struct *work)
1542 {
1543 	struct delayed_work *delay = to_delayed_work(work);
1544 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1545 						 service_task);
1546 	struct mlx4_en_dev *mdev = priv->mdev;
1547 
1548 	mutex_lock(&mdev->state_lock);
1549 	if (mdev->device_up) {
1550 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1551 			mlx4_en_ptp_overflow_check(mdev);
1552 
1553 		mlx4_en_recover_from_oom(priv);
1554 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1555 				   SERVICE_TASK_DELAY);
1556 	}
1557 	mutex_unlock(&mdev->state_lock);
1558 }
1559 
1560 static void mlx4_en_linkstate(struct work_struct *work)
1561 {
1562 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1563 						 linkstate_task);
1564 	struct mlx4_en_dev *mdev = priv->mdev;
1565 	int linkstate = priv->link_state;
1566 
1567 	mutex_lock(&mdev->state_lock);
1568 	/* If observable port state changed set carrier state and
1569 	 * report to system log */
1570 	if (priv->last_link_state != linkstate) {
1571 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1572 			en_info(priv, "Link Down\n");
1573 			netif_carrier_off(priv->dev);
1574 		} else {
1575 			en_info(priv, "Link Up\n");
1576 			netif_carrier_on(priv->dev);
1577 		}
1578 	}
1579 	priv->last_link_state = linkstate;
1580 	mutex_unlock(&mdev->state_lock);
1581 }
1582 
1583 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1584 {
1585 	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1586 	int numa_node = priv->mdev->dev->numa_node;
1587 
1588 	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1589 		return -ENOMEM;
1590 
1591 	cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1592 			ring->affinity_mask);
1593 	return 0;
1594 }
1595 
1596 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1597 {
1598 	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1599 }
1600 
1601 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1602 				      int tx_ring_idx)
1603 {
1604 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1605 	int rr_index = tx_ring_idx;
1606 
1607 	tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1608 	tx_ring->recycle_ring = priv->rx_ring[rr_index];
1609 	en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1610 	       TX_XDP, tx_ring_idx, rr_index);
1611 }
1612 
1613 int mlx4_en_start_port(struct net_device *dev)
1614 {
1615 	struct mlx4_en_priv *priv = netdev_priv(dev);
1616 	struct mlx4_en_dev *mdev = priv->mdev;
1617 	struct mlx4_en_cq *cq;
1618 	struct mlx4_en_tx_ring *tx_ring;
1619 	int rx_index = 0;
1620 	int err = 0;
1621 	int i, t;
1622 	int j;
1623 	u8 mc_list[16] = {0};
1624 
1625 	if (priv->port_up) {
1626 		en_dbg(DRV, priv, "start port called while port already up\n");
1627 		return 0;
1628 	}
1629 
1630 	INIT_LIST_HEAD(&priv->mc_list);
1631 	INIT_LIST_HEAD(&priv->curr_list);
1632 	INIT_LIST_HEAD(&priv->ethtool_list);
1633 	memset(&priv->ethtool_rules[0], 0,
1634 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1635 
1636 	/* Calculate Rx buf size */
1637 	dev->mtu = min(dev->mtu, priv->max_mtu);
1638 	mlx4_en_calc_rx_buf(dev);
1639 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1640 
1641 	/* Configure rx cq's and rings */
1642 	err = mlx4_en_activate_rx_rings(priv);
1643 	if (err) {
1644 		en_err(priv, "Failed to activate RX rings\n");
1645 		return err;
1646 	}
1647 	for (i = 0; i < priv->rx_ring_num; i++) {
1648 		cq = priv->rx_cq[i];
1649 
1650 		err = mlx4_en_init_affinity_hint(priv, i);
1651 		if (err) {
1652 			en_err(priv, "Failed preparing IRQ affinity hint\n");
1653 			goto cq_err;
1654 		}
1655 
1656 		err = mlx4_en_activate_cq(priv, cq, i);
1657 		if (err) {
1658 			en_err(priv, "Failed activating Rx CQ\n");
1659 			mlx4_en_free_affinity_hint(priv, i);
1660 			goto cq_err;
1661 		}
1662 
1663 		for (j = 0; j < cq->size; j++) {
1664 			struct mlx4_cqe *cqe = NULL;
1665 
1666 			cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1667 			      priv->cqe_factor;
1668 			cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1669 		}
1670 
1671 		err = mlx4_en_set_cq_moder(priv, cq);
1672 		if (err) {
1673 			en_err(priv, "Failed setting cq moderation parameters\n");
1674 			mlx4_en_deactivate_cq(priv, cq);
1675 			mlx4_en_free_affinity_hint(priv, i);
1676 			goto cq_err;
1677 		}
1678 		mlx4_en_arm_cq(priv, cq);
1679 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1680 		++rx_index;
1681 	}
1682 
1683 	/* Set qp number */
1684 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1685 	err = mlx4_en_get_qp(priv);
1686 	if (err) {
1687 		en_err(priv, "Failed getting eth qp\n");
1688 		goto cq_err;
1689 	}
1690 	mdev->mac_removed[priv->port] = 0;
1691 
1692 	priv->counter_index =
1693 			mlx4_get_default_counter_index(mdev->dev, priv->port);
1694 
1695 	err = mlx4_en_config_rss_steer(priv);
1696 	if (err) {
1697 		en_err(priv, "Failed configuring rss steering\n");
1698 		goto mac_err;
1699 	}
1700 
1701 	err = mlx4_en_create_drop_qp(priv);
1702 	if (err)
1703 		goto rss_err;
1704 
1705 	/* Configure tx cq's and rings */
1706 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1707 		u8 num_tx_rings_p_up = t == TX ?
1708 			priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1709 
1710 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1711 			/* Configure cq */
1712 			cq = priv->tx_cq[t][i];
1713 			err = mlx4_en_activate_cq(priv, cq, i);
1714 			if (err) {
1715 				en_err(priv, "Failed allocating Tx CQ\n");
1716 				goto tx_err;
1717 			}
1718 			err = mlx4_en_set_cq_moder(priv, cq);
1719 			if (err) {
1720 				en_err(priv, "Failed setting cq moderation parameters\n");
1721 				mlx4_en_deactivate_cq(priv, cq);
1722 				goto tx_err;
1723 			}
1724 			en_dbg(DRV, priv,
1725 			       "Resetting index of collapsed CQ:%d to -1\n", i);
1726 			cq->buf->wqe_index = cpu_to_be16(0xffff);
1727 
1728 			/* Configure ring */
1729 			tx_ring = priv->tx_ring[t][i];
1730 			err = mlx4_en_activate_tx_ring(priv, tx_ring,
1731 						       cq->mcq.cqn,
1732 						       i / num_tx_rings_p_up);
1733 			if (err) {
1734 				en_err(priv, "Failed allocating Tx ring\n");
1735 				mlx4_en_deactivate_cq(priv, cq);
1736 				goto tx_err;
1737 			}
1738 			clear_bit(MLX4_EN_TX_RING_STATE_RECOVERING, &tx_ring->state);
1739 			if (t != TX_XDP) {
1740 				tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1741 				tx_ring->recycle_ring = NULL;
1742 
1743 				/* Arm CQ for TX completions */
1744 				mlx4_en_arm_cq(priv, cq);
1745 
1746 			} else {
1747 				mlx4_en_init_tx_xdp_ring_descs(priv, tx_ring);
1748 				mlx4_en_init_recycle_ring(priv, i);
1749 				/* XDP TX CQ should never be armed */
1750 			}
1751 
1752 			/* Set initial ownership of all Tx TXBBs to SW (1) */
1753 			for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1754 				*((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1755 		}
1756 	}
1757 
1758 	/* Configure port */
1759 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1760 				    priv->rx_skb_size + ETH_FCS_LEN,
1761 				    priv->prof->tx_pause,
1762 				    priv->prof->tx_ppp,
1763 				    priv->prof->rx_pause,
1764 				    priv->prof->rx_ppp);
1765 	if (err) {
1766 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1767 		       priv->port, err);
1768 		goto tx_err;
1769 	}
1770 
1771 	err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1772 	if (err) {
1773 		en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1774 		       dev->mtu, priv->port, err);
1775 		goto tx_err;
1776 	}
1777 
1778 	/* Set default qp number */
1779 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1780 	if (err) {
1781 		en_err(priv, "Failed setting default qp numbers\n");
1782 		goto tx_err;
1783 	}
1784 
1785 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1786 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1787 		if (err) {
1788 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1789 			       err);
1790 			goto tx_err;
1791 		}
1792 	}
1793 
1794 	/* Init port */
1795 	en_dbg(HW, priv, "Initializing port\n");
1796 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1797 	if (err) {
1798 		en_err(priv, "Failed Initializing port\n");
1799 		goto tx_err;
1800 	}
1801 
1802 	/* Set Unicast and VXLAN steering rules */
1803 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1804 	    mlx4_en_set_rss_steer_rules(priv))
1805 		mlx4_warn(mdev, "Failed setting steering rules\n");
1806 
1807 	/* Attach rx QP to bradcast address */
1808 	eth_broadcast_addr(&mc_list[10]);
1809 	mc_list[5] = priv->port; /* needed for B0 steering support */
1810 	if (mlx4_multicast_attach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1811 				  priv->port, 0, MLX4_PROT_ETH,
1812 				  &priv->broadcast_id))
1813 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1814 
1815 	/* Must redo promiscuous mode setup. */
1816 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1817 
1818 	/* Schedule multicast task to populate multicast list */
1819 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1820 
1821 	if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1822 		udp_tunnel_nic_reset_ntf(dev);
1823 
1824 	priv->port_up = true;
1825 
1826 	/* Process all completions if exist to prevent
1827 	 * the queues freezing if they are full
1828 	 */
1829 	for (i = 0; i < priv->rx_ring_num; i++) {
1830 		local_bh_disable();
1831 		napi_schedule(&priv->rx_cq[i]->napi);
1832 		local_bh_enable();
1833 	}
1834 
1835 	clear_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state);
1836 	netif_tx_start_all_queues(dev);
1837 	netif_device_attach(dev);
1838 
1839 	return 0;
1840 
1841 tx_err:
1842 	if (t == MLX4_EN_NUM_TX_TYPES) {
1843 		t--;
1844 		i = priv->tx_ring_num[t];
1845 	}
1846 	while (t >= 0) {
1847 		while (i--) {
1848 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1849 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1850 		}
1851 		if (!t--)
1852 			break;
1853 		i = priv->tx_ring_num[t];
1854 	}
1855 	mlx4_en_destroy_drop_qp(priv);
1856 rss_err:
1857 	mlx4_en_release_rss_steer(priv);
1858 mac_err:
1859 	mlx4_en_put_qp(priv);
1860 cq_err:
1861 	while (rx_index--) {
1862 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1863 		mlx4_en_free_affinity_hint(priv, rx_index);
1864 	}
1865 	for (i = 0; i < priv->rx_ring_num; i++)
1866 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1867 
1868 	return err; /* need to close devices */
1869 }
1870 
1871 
1872 void mlx4_en_stop_port(struct net_device *dev, int detach)
1873 {
1874 	struct mlx4_en_priv *priv = netdev_priv(dev);
1875 	struct mlx4_en_dev *mdev = priv->mdev;
1876 	struct mlx4_en_mc_list *mclist, *tmp;
1877 	struct ethtool_flow_id *flow, *tmp_flow;
1878 	int i, t;
1879 	u8 mc_list[16] = {0};
1880 
1881 	if (!priv->port_up) {
1882 		en_dbg(DRV, priv, "stop port called while port already down\n");
1883 		return;
1884 	}
1885 
1886 	/* close port*/
1887 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1888 
1889 	/* Synchronize with tx routine */
1890 	netif_tx_lock_bh(dev);
1891 	if (detach)
1892 		netif_device_detach(dev);
1893 	netif_tx_stop_all_queues(dev);
1894 	netif_tx_unlock_bh(dev);
1895 
1896 	netif_tx_disable(dev);
1897 
1898 	spin_lock_bh(&priv->stats_lock);
1899 	mlx4_en_fold_software_stats(dev);
1900 	/* Set port as not active */
1901 	priv->port_up = false;
1902 	spin_unlock_bh(&priv->stats_lock);
1903 
1904 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1905 
1906 	/* Promsicuous mode */
1907 	if (mdev->dev->caps.steering_mode ==
1908 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1909 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1910 				 MLX4_EN_FLAG_MC_PROMISC);
1911 		mlx4_flow_steer_promisc_remove(mdev->dev,
1912 					       priv->port,
1913 					       MLX4_FS_ALL_DEFAULT);
1914 		mlx4_flow_steer_promisc_remove(mdev->dev,
1915 					       priv->port,
1916 					       MLX4_FS_MC_DEFAULT);
1917 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1918 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1919 
1920 		/* Disable promiscouos mode */
1921 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1922 					    priv->port);
1923 
1924 		/* Disable Multicast promisc */
1925 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1926 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1927 						      priv->port);
1928 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1929 		}
1930 	}
1931 
1932 	/* Detach All multicasts */
1933 	eth_broadcast_addr(&mc_list[10]);
1934 	mc_list[5] = priv->port; /* needed for B0 steering support */
1935 	mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1936 			      MLX4_PROT_ETH, priv->broadcast_id);
1937 	list_for_each_entry(mclist, &priv->curr_list, list) {
1938 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1939 		mc_list[5] = priv->port;
1940 		mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp,
1941 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1942 		if (mclist->tunnel_reg_id)
1943 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1944 	}
1945 	mlx4_en_clear_list(dev);
1946 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1947 		list_del(&mclist->list);
1948 		kfree(mclist);
1949 	}
1950 
1951 	/* Flush multicast filter */
1952 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1953 
1954 	/* Remove flow steering rules for the port*/
1955 	if (mdev->dev->caps.steering_mode ==
1956 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1957 		ASSERT_RTNL();
1958 		list_for_each_entry_safe(flow, tmp_flow,
1959 					 &priv->ethtool_list, list) {
1960 			mlx4_flow_detach(mdev->dev, flow->id);
1961 			list_del(&flow->list);
1962 		}
1963 	}
1964 
1965 	mlx4_en_destroy_drop_qp(priv);
1966 
1967 	/* Free TX Rings */
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_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1971 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1972 		}
1973 	}
1974 	msleep(10);
1975 
1976 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1977 		for (i = 0; i < priv->tx_ring_num[t]; i++)
1978 			mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1979 
1980 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1981 		mlx4_en_delete_rss_steer_rules(priv);
1982 
1983 	/* Free RSS qps */
1984 	mlx4_en_release_rss_steer(priv);
1985 
1986 	/* Unregister Mac address for the port */
1987 	mlx4_en_put_qp(priv);
1988 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1989 		mdev->mac_removed[priv->port] = 1;
1990 
1991 	/* Free RX Rings */
1992 	for (i = 0; i < priv->rx_ring_num; i++) {
1993 		struct mlx4_en_cq *cq = priv->rx_cq[i];
1994 
1995 		napi_synchronize(&cq->napi);
1996 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1997 		mlx4_en_deactivate_cq(priv, cq);
1998 
1999 		mlx4_en_free_affinity_hint(priv, i);
2000 	}
2001 }
2002 
2003 static void mlx4_en_restart(struct work_struct *work)
2004 {
2005 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2006 						 restart_task);
2007 	struct mlx4_en_dev *mdev = priv->mdev;
2008 	struct net_device *dev = priv->dev;
2009 
2010 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
2011 
2012 	rtnl_lock();
2013 	mutex_lock(&mdev->state_lock);
2014 	if (priv->port_up) {
2015 		mlx4_en_stop_port(dev, 1);
2016 		if (mlx4_en_start_port(dev))
2017 			en_err(priv, "Failed restarting port %d\n", priv->port);
2018 	}
2019 	mutex_unlock(&mdev->state_lock);
2020 	rtnl_unlock();
2021 }
2022 
2023 static void mlx4_en_clear_stats(struct net_device *dev)
2024 {
2025 	struct mlx4_en_priv *priv = netdev_priv(dev);
2026 	struct mlx4_en_dev *mdev = priv->mdev;
2027 	struct mlx4_en_tx_ring **tx_ring;
2028 	int i;
2029 
2030 	if (!mlx4_is_slave(mdev->dev))
2031 		if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
2032 			en_dbg(HW, priv, "Failed dumping statistics\n");
2033 
2034 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
2035 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
2036 	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
2037 	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
2038 	memset(&priv->rx_priority_flowstats, 0,
2039 	       sizeof(priv->rx_priority_flowstats));
2040 	memset(&priv->tx_priority_flowstats, 0,
2041 	       sizeof(priv->tx_priority_flowstats));
2042 	memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
2043 
2044 	tx_ring = priv->tx_ring[TX];
2045 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
2046 		tx_ring[i]->bytes = 0;
2047 		tx_ring[i]->packets = 0;
2048 		tx_ring[i]->tx_csum = 0;
2049 		tx_ring[i]->tx_dropped = 0;
2050 		tx_ring[i]->queue_stopped = 0;
2051 		tx_ring[i]->wake_queue = 0;
2052 		tx_ring[i]->tso_packets = 0;
2053 		tx_ring[i]->xmit_more = 0;
2054 	}
2055 	for (i = 0; i < priv->rx_ring_num; i++) {
2056 		priv->rx_ring[i]->bytes = 0;
2057 		priv->rx_ring[i]->packets = 0;
2058 		priv->rx_ring[i]->csum_ok = 0;
2059 		priv->rx_ring[i]->csum_none = 0;
2060 		priv->rx_ring[i]->csum_complete = 0;
2061 	}
2062 }
2063 
2064 static int mlx4_en_open(struct net_device *dev)
2065 {
2066 	struct mlx4_en_priv *priv = netdev_priv(dev);
2067 	struct mlx4_en_dev *mdev = priv->mdev;
2068 	int err = 0;
2069 
2070 	mutex_lock(&mdev->state_lock);
2071 
2072 	if (!mdev->device_up) {
2073 		en_err(priv, "Cannot open - device down/disabled\n");
2074 		err = -EBUSY;
2075 		goto out;
2076 	}
2077 
2078 	/* Reset HW statistics and SW counters */
2079 	mlx4_en_clear_stats(dev);
2080 
2081 	err = mlx4_en_start_port(dev);
2082 	if (err)
2083 		en_err(priv, "Failed starting port:%d\n", priv->port);
2084 
2085 out:
2086 	mutex_unlock(&mdev->state_lock);
2087 	return err;
2088 }
2089 
2090 
2091 static int mlx4_en_close(struct net_device *dev)
2092 {
2093 	struct mlx4_en_priv *priv = netdev_priv(dev);
2094 	struct mlx4_en_dev *mdev = priv->mdev;
2095 
2096 	en_dbg(IFDOWN, priv, "Close port called\n");
2097 
2098 	mutex_lock(&mdev->state_lock);
2099 
2100 	mlx4_en_stop_port(dev, 0);
2101 	netif_carrier_off(dev);
2102 
2103 	mutex_unlock(&mdev->state_lock);
2104 	return 0;
2105 }
2106 
2107 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2108 {
2109 	int i, t;
2110 
2111 #ifdef CONFIG_RFS_ACCEL
2112 	priv->dev->rx_cpu_rmap = NULL;
2113 #endif
2114 
2115 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2116 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2117 			if (priv->tx_ring[t] && priv->tx_ring[t][i])
2118 				mlx4_en_destroy_tx_ring(priv,
2119 							&priv->tx_ring[t][i]);
2120 			if (priv->tx_cq[t] && priv->tx_cq[t][i])
2121 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2122 		}
2123 		kfree(priv->tx_ring[t]);
2124 		kfree(priv->tx_cq[t]);
2125 	}
2126 
2127 	for (i = 0; i < priv->rx_ring_num; i++) {
2128 		if (priv->rx_ring[i])
2129 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2130 				priv->prof->rx_ring_size, priv->stride);
2131 		if (priv->rx_cq[i])
2132 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2133 	}
2134 
2135 }
2136 
2137 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2138 {
2139 	struct mlx4_en_port_profile *prof = priv->prof;
2140 	int i, t;
2141 	int node;
2142 
2143 	/* Create tx Rings */
2144 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2145 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2146 			node = cpu_to_node(i % num_online_cpus());
2147 			if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2148 					      prof->tx_ring_size, i, t, node))
2149 				goto err;
2150 
2151 			if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2152 						   prof->tx_ring_size,
2153 						   TXBB_SIZE, node, i))
2154 				goto err;
2155 		}
2156 	}
2157 
2158 	/* Create rx Rings */
2159 	for (i = 0; i < priv->rx_ring_num; i++) {
2160 		node = cpu_to_node(i % num_online_cpus());
2161 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2162 				      prof->rx_ring_size, i, RX, node))
2163 			goto err;
2164 
2165 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2166 					   prof->rx_ring_size, priv->stride,
2167 					   node, i))
2168 			goto err;
2169 
2170 	}
2171 
2172 #ifdef CONFIG_RFS_ACCEL
2173 	priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2174 #endif
2175 
2176 	return 0;
2177 
2178 err:
2179 	en_err(priv, "Failed to allocate NIC resources\n");
2180 	for (i = 0; i < priv->rx_ring_num; i++) {
2181 		if (priv->rx_ring[i])
2182 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2183 						prof->rx_ring_size,
2184 						priv->stride);
2185 		if (priv->rx_cq[i])
2186 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2187 	}
2188 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2189 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2190 			if (priv->tx_ring[t][i])
2191 				mlx4_en_destroy_tx_ring(priv,
2192 							&priv->tx_ring[t][i]);
2193 			if (priv->tx_cq[t][i])
2194 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2195 		}
2196 	}
2197 	return -ENOMEM;
2198 }
2199 
2200 
2201 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2202 			     struct mlx4_en_priv *src,
2203 			     struct mlx4_en_port_profile *prof)
2204 {
2205 	int t;
2206 
2207 	memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2208 	       sizeof(dst->hwtstamp_config));
2209 	dst->num_tx_rings_p_up = prof->num_tx_rings_p_up;
2210 	dst->rx_ring_num = prof->rx_ring_num;
2211 	dst->flags = prof->flags;
2212 	dst->mdev = src->mdev;
2213 	dst->port = src->port;
2214 	dst->dev = src->dev;
2215 	dst->prof = prof;
2216 	dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2217 					 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2218 
2219 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2220 		dst->tx_ring_num[t] = prof->tx_ring_num[t];
2221 		if (!dst->tx_ring_num[t])
2222 			continue;
2223 
2224 		dst->tx_ring[t] = kcalloc(MAX_TX_RINGS,
2225 					  sizeof(struct mlx4_en_tx_ring *),
2226 					  GFP_KERNEL);
2227 		if (!dst->tx_ring[t])
2228 			goto err_free_tx;
2229 
2230 		dst->tx_cq[t] = kcalloc(MAX_TX_RINGS,
2231 					sizeof(struct mlx4_en_cq *),
2232 					GFP_KERNEL);
2233 		if (!dst->tx_cq[t]) {
2234 			kfree(dst->tx_ring[t]);
2235 			goto err_free_tx;
2236 		}
2237 	}
2238 
2239 	return 0;
2240 
2241 err_free_tx:
2242 	while (t--) {
2243 		kfree(dst->tx_ring[t]);
2244 		kfree(dst->tx_cq[t]);
2245 	}
2246 	return -ENOMEM;
2247 }
2248 
2249 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2250 				struct mlx4_en_priv *src)
2251 {
2252 	int t;
2253 	memcpy(dst->rx_ring, src->rx_ring,
2254 	       sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2255 	memcpy(dst->rx_cq, src->rx_cq,
2256 	       sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2257 	memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2258 	       sizeof(dst->hwtstamp_config));
2259 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2260 		dst->tx_ring_num[t] = src->tx_ring_num[t];
2261 		dst->tx_ring[t] = src->tx_ring[t];
2262 		dst->tx_cq[t] = src->tx_cq[t];
2263 	}
2264 	dst->num_tx_rings_p_up = src->num_tx_rings_p_up;
2265 	dst->rx_ring_num = src->rx_ring_num;
2266 	memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2267 }
2268 
2269 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2270 				struct mlx4_en_priv *tmp,
2271 				struct mlx4_en_port_profile *prof,
2272 				bool carry_xdp_prog)
2273 {
2274 	struct bpf_prog *xdp_prog;
2275 	int i, t;
2276 
2277 	mlx4_en_copy_priv(tmp, priv, prof);
2278 
2279 	if (mlx4_en_alloc_resources(tmp)) {
2280 		en_warn(priv,
2281 			"%s: Resource allocation failed, using previous configuration\n",
2282 			__func__);
2283 		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2284 			kfree(tmp->tx_ring[t]);
2285 			kfree(tmp->tx_cq[t]);
2286 		}
2287 		return -ENOMEM;
2288 	}
2289 
2290 	/* All rx_rings has the same xdp_prog.  Pick the first one. */
2291 	xdp_prog = rcu_dereference_protected(
2292 		priv->rx_ring[0]->xdp_prog,
2293 		lockdep_is_held(&priv->mdev->state_lock));
2294 
2295 	if (xdp_prog && carry_xdp_prog) {
2296 		bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2297 		for (i = 0; i < tmp->rx_ring_num; i++)
2298 			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2299 					   xdp_prog);
2300 	}
2301 
2302 	return 0;
2303 }
2304 
2305 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2306 				    struct mlx4_en_priv *tmp)
2307 {
2308 	mlx4_en_free_resources(priv);
2309 	mlx4_en_update_priv(priv, tmp);
2310 }
2311 
2312 void mlx4_en_destroy_netdev(struct net_device *dev)
2313 {
2314 	struct mlx4_en_priv *priv = netdev_priv(dev);
2315 	struct mlx4_en_dev *mdev = priv->mdev;
2316 
2317 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2318 
2319 	/* Unregister device - this will close the port if it was up */
2320 	if (priv->registered) {
2321 		devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2322 							      priv->port));
2323 		unregister_netdev(dev);
2324 	}
2325 
2326 	if (priv->allocated)
2327 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2328 
2329 	cancel_delayed_work(&priv->stats_task);
2330 	cancel_delayed_work(&priv->service_task);
2331 	/* flush any pending task for this netdev */
2332 	flush_workqueue(mdev->workqueue);
2333 
2334 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2335 		mlx4_en_remove_timestamp(mdev);
2336 
2337 	/* Detach the netdev so tasks would not attempt to access it */
2338 	mutex_lock(&mdev->state_lock);
2339 	mdev->pndev[priv->port] = NULL;
2340 	mdev->upper[priv->port] = NULL;
2341 
2342 #ifdef CONFIG_RFS_ACCEL
2343 	mlx4_en_cleanup_filters(priv);
2344 #endif
2345 
2346 	mlx4_en_free_resources(priv);
2347 	mutex_unlock(&mdev->state_lock);
2348 
2349 	free_netdev(dev);
2350 }
2351 
2352 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2353 {
2354 	struct mlx4_en_priv *priv = netdev_priv(dev);
2355 
2356 	if (mtu > MLX4_EN_MAX_XDP_MTU) {
2357 		en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2358 		       mtu, MLX4_EN_MAX_XDP_MTU);
2359 		return false;
2360 	}
2361 
2362 	return true;
2363 }
2364 
2365 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2366 {
2367 	struct mlx4_en_priv *priv = netdev_priv(dev);
2368 	struct mlx4_en_dev *mdev = priv->mdev;
2369 	int err = 0;
2370 
2371 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2372 		 dev->mtu, new_mtu);
2373 
2374 	if (priv->tx_ring_num[TX_XDP] &&
2375 	    !mlx4_en_check_xdp_mtu(dev, new_mtu))
2376 		return -EOPNOTSUPP;
2377 
2378 	dev->mtu = new_mtu;
2379 
2380 	if (netif_running(dev)) {
2381 		mutex_lock(&mdev->state_lock);
2382 		if (!mdev->device_up) {
2383 			/* NIC is probably restarting - let restart task reset
2384 			 * the port */
2385 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2386 		} else {
2387 			mlx4_en_stop_port(dev, 1);
2388 			err = mlx4_en_start_port(dev);
2389 			if (err) {
2390 				en_err(priv, "Failed restarting port:%d\n",
2391 					 priv->port);
2392 				if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING,
2393 						      &priv->state))
2394 					queue_work(mdev->workqueue, &priv->restart_task);
2395 			}
2396 		}
2397 		mutex_unlock(&mdev->state_lock);
2398 	}
2399 	return 0;
2400 }
2401 
2402 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2403 {
2404 	struct mlx4_en_priv *priv = netdev_priv(dev);
2405 	struct mlx4_en_dev *mdev = priv->mdev;
2406 	struct hwtstamp_config config;
2407 
2408 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2409 		return -EFAULT;
2410 
2411 	/* reserved for future extensions */
2412 	if (config.flags)
2413 		return -EINVAL;
2414 
2415 	/* device doesn't support time stamping */
2416 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2417 		return -EINVAL;
2418 
2419 	/* TX HW timestamp */
2420 	switch (config.tx_type) {
2421 	case HWTSTAMP_TX_OFF:
2422 	case HWTSTAMP_TX_ON:
2423 		break;
2424 	default:
2425 		return -ERANGE;
2426 	}
2427 
2428 	/* RX HW timestamp */
2429 	switch (config.rx_filter) {
2430 	case HWTSTAMP_FILTER_NONE:
2431 		break;
2432 	case HWTSTAMP_FILTER_ALL:
2433 	case HWTSTAMP_FILTER_SOME:
2434 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2435 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2436 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2437 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2438 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2439 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2440 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2441 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2442 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2443 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2444 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2445 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2446 	case HWTSTAMP_FILTER_NTP_ALL:
2447 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2448 		break;
2449 	default:
2450 		return -ERANGE;
2451 	}
2452 
2453 	if (mlx4_en_reset_config(dev, config, dev->features)) {
2454 		config.tx_type = HWTSTAMP_TX_OFF;
2455 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2456 	}
2457 
2458 	return copy_to_user(ifr->ifr_data, &config,
2459 			    sizeof(config)) ? -EFAULT : 0;
2460 }
2461 
2462 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2463 {
2464 	struct mlx4_en_priv *priv = netdev_priv(dev);
2465 
2466 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2467 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2468 }
2469 
2470 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2471 {
2472 	switch (cmd) {
2473 	case SIOCSHWTSTAMP:
2474 		return mlx4_en_hwtstamp_set(dev, ifr);
2475 	case SIOCGHWTSTAMP:
2476 		return mlx4_en_hwtstamp_get(dev, ifr);
2477 	default:
2478 		return -EOPNOTSUPP;
2479 	}
2480 }
2481 
2482 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2483 					      netdev_features_t features)
2484 {
2485 	struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2486 	struct mlx4_en_dev *mdev = en_priv->mdev;
2487 
2488 	/* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2489 	 * enable/disable make sure S-TAG flag is always in same state as
2490 	 * C-TAG.
2491 	 */
2492 	if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2493 	    !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2494 		features |= NETIF_F_HW_VLAN_STAG_RX;
2495 	else
2496 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2497 
2498 	return features;
2499 }
2500 
2501 static int mlx4_en_set_features(struct net_device *netdev,
2502 		netdev_features_t features)
2503 {
2504 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2505 	bool reset = false;
2506 	int ret = 0;
2507 
2508 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2509 		en_info(priv, "Turn %s RX-FCS\n",
2510 			(features & NETIF_F_RXFCS) ? "ON" : "OFF");
2511 		reset = true;
2512 	}
2513 
2514 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2515 		u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2516 
2517 		en_info(priv, "Turn %s RX-ALL\n",
2518 			ignore_fcs_value ? "ON" : "OFF");
2519 		ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2520 					      priv->port, ignore_fcs_value);
2521 		if (ret)
2522 			return ret;
2523 	}
2524 
2525 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2526 		en_info(priv, "Turn %s RX vlan strip offload\n",
2527 			(features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2528 		reset = true;
2529 	}
2530 
2531 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2532 		en_info(priv, "Turn %s TX vlan strip offload\n",
2533 			(features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2534 
2535 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2536 		en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2537 			(features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2538 
2539 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2540 		en_info(priv, "Turn %s loopback\n",
2541 			(features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2542 		mlx4_en_update_loopback_state(netdev, features);
2543 	}
2544 
2545 	if (reset) {
2546 		ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2547 					   features);
2548 		if (ret)
2549 			return ret;
2550 	}
2551 
2552 	return 0;
2553 }
2554 
2555 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2556 {
2557 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2558 	struct mlx4_en_dev *mdev = en_priv->mdev;
2559 
2560 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac);
2561 }
2562 
2563 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2564 			       __be16 vlan_proto)
2565 {
2566 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2567 	struct mlx4_en_dev *mdev = en_priv->mdev;
2568 
2569 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2570 				vlan_proto);
2571 }
2572 
2573 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2574 			       int max_tx_rate)
2575 {
2576 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2577 	struct mlx4_en_dev *mdev = en_priv->mdev;
2578 
2579 	return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2580 				max_tx_rate);
2581 }
2582 
2583 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2584 {
2585 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2586 	struct mlx4_en_dev *mdev = en_priv->mdev;
2587 
2588 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2589 }
2590 
2591 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2592 {
2593 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2594 	struct mlx4_en_dev *mdev = en_priv->mdev;
2595 
2596 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2597 }
2598 
2599 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2600 {
2601 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2602 	struct mlx4_en_dev *mdev = en_priv->mdev;
2603 
2604 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2605 }
2606 
2607 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2608 				struct ifla_vf_stats *vf_stats)
2609 {
2610 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2611 	struct mlx4_en_dev *mdev = en_priv->mdev;
2612 
2613 	return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2614 }
2615 
2616 #define PORT_ID_BYTE_LEN 8
2617 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2618 				    struct netdev_phys_item_id *ppid)
2619 {
2620 	struct mlx4_en_priv *priv = netdev_priv(dev);
2621 	struct mlx4_dev *mdev = priv->mdev->dev;
2622 	int i;
2623 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2624 
2625 	if (!phys_port_id)
2626 		return -EOPNOTSUPP;
2627 
2628 	ppid->id_len = sizeof(phys_port_id);
2629 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2630 		ppid->id[i] =  phys_port_id & 0xff;
2631 		phys_port_id >>= 8;
2632 	}
2633 	return 0;
2634 }
2635 
2636 static int mlx4_udp_tunnel_sync(struct net_device *dev, unsigned int table)
2637 {
2638 	struct mlx4_en_priv *priv = netdev_priv(dev);
2639 	struct udp_tunnel_info ti;
2640 	int ret;
2641 
2642 	udp_tunnel_nic_get_port(dev, table, 0, &ti);
2643 	priv->vxlan_port = ti.port;
2644 
2645 	ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2646 	if (ret)
2647 		return ret;
2648 
2649 	return mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2650 				   VXLAN_STEER_BY_OUTER_MAC,
2651 				   !!priv->vxlan_port);
2652 }
2653 
2654 static const struct udp_tunnel_nic_info mlx4_udp_tunnels = {
2655 	.sync_table	= mlx4_udp_tunnel_sync,
2656 	.flags		= UDP_TUNNEL_NIC_INFO_MAY_SLEEP |
2657 			  UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
2658 	.tables		= {
2659 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
2660 	},
2661 };
2662 
2663 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2664 						struct net_device *dev,
2665 						netdev_features_t features)
2666 {
2667 	features = vlan_features_check(skb, features);
2668 	features = vxlan_features_check(skb, features);
2669 
2670 	/* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2671 	 * support inner IPv6 checksums and segmentation so  we need to
2672 	 * strip that feature if this is an IPv6 encapsulated frame.
2673 	 */
2674 	if (skb->encapsulation &&
2675 	    (skb->ip_summed == CHECKSUM_PARTIAL)) {
2676 		struct mlx4_en_priv *priv = netdev_priv(dev);
2677 
2678 		if (!priv->vxlan_port ||
2679 		    (ip_hdr(skb)->version != 4) ||
2680 		    (udp_hdr(skb)->dest != priv->vxlan_port))
2681 			features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2682 	}
2683 
2684 	return features;
2685 }
2686 
2687 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2688 {
2689 	struct mlx4_en_priv *priv = netdev_priv(dev);
2690 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2691 	struct mlx4_update_qp_params params;
2692 	int err;
2693 
2694 	if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2695 		return -EOPNOTSUPP;
2696 
2697 	/* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2698 	if (maxrate >> 12) {
2699 		params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2700 		params.rate_val  = maxrate / 1000;
2701 	} else if (maxrate) {
2702 		params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2703 		params.rate_val  = maxrate;
2704 	} else { /* zero serves to revoke the QP rate-limitation */
2705 		params.rate_unit = 0;
2706 		params.rate_val  = 0;
2707 	}
2708 
2709 	err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2710 			     &params);
2711 	return err;
2712 }
2713 
2714 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2715 {
2716 	struct mlx4_en_priv *priv = netdev_priv(dev);
2717 	struct mlx4_en_dev *mdev = priv->mdev;
2718 	struct mlx4_en_port_profile new_prof;
2719 	struct bpf_prog *old_prog;
2720 	struct mlx4_en_priv *tmp;
2721 	int tx_changed = 0;
2722 	int xdp_ring_num;
2723 	int port_up = 0;
2724 	int err;
2725 	int i;
2726 
2727 	xdp_ring_num = prog ? priv->rx_ring_num : 0;
2728 
2729 	/* No need to reconfigure buffers when simply swapping the
2730 	 * program for a new one.
2731 	 */
2732 	if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2733 		if (prog)
2734 			bpf_prog_add(prog, priv->rx_ring_num - 1);
2735 
2736 		mutex_lock(&mdev->state_lock);
2737 		for (i = 0; i < priv->rx_ring_num; i++) {
2738 			old_prog = rcu_dereference_protected(
2739 					priv->rx_ring[i]->xdp_prog,
2740 					lockdep_is_held(&mdev->state_lock));
2741 			rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2742 			if (old_prog)
2743 				bpf_prog_put(old_prog);
2744 		}
2745 		mutex_unlock(&mdev->state_lock);
2746 		return 0;
2747 	}
2748 
2749 	if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2750 		return -EOPNOTSUPP;
2751 
2752 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2753 	if (!tmp)
2754 		return -ENOMEM;
2755 
2756 	if (prog)
2757 		bpf_prog_add(prog, priv->rx_ring_num - 1);
2758 
2759 	mutex_lock(&mdev->state_lock);
2760 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2761 	new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2762 
2763 	if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2764 		tx_changed = 1;
2765 		new_prof.tx_ring_num[TX] =
2766 			MAX_TX_RINGS - ALIGN(xdp_ring_num, priv->prof->num_up);
2767 		en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2768 	}
2769 
2770 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2771 	if (err) {
2772 		if (prog)
2773 			bpf_prog_sub(prog, priv->rx_ring_num - 1);
2774 		goto unlock_out;
2775 	}
2776 
2777 	if (priv->port_up) {
2778 		port_up = 1;
2779 		mlx4_en_stop_port(dev, 1);
2780 	}
2781 
2782 	mlx4_en_safe_replace_resources(priv, tmp);
2783 	if (tx_changed)
2784 		netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2785 
2786 	for (i = 0; i < priv->rx_ring_num; i++) {
2787 		old_prog = rcu_dereference_protected(
2788 					priv->rx_ring[i]->xdp_prog,
2789 					lockdep_is_held(&mdev->state_lock));
2790 		rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2791 		if (old_prog)
2792 			bpf_prog_put(old_prog);
2793 	}
2794 
2795 	if (port_up) {
2796 		err = mlx4_en_start_port(dev);
2797 		if (err) {
2798 			en_err(priv, "Failed starting port %d for XDP change\n",
2799 			       priv->port);
2800 			if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state))
2801 				queue_work(mdev->workqueue, &priv->restart_task);
2802 		}
2803 	}
2804 
2805 unlock_out:
2806 	mutex_unlock(&mdev->state_lock);
2807 	kfree(tmp);
2808 	return err;
2809 }
2810 
2811 static int mlx4_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2812 {
2813 	switch (xdp->command) {
2814 	case XDP_SETUP_PROG:
2815 		return mlx4_xdp_set(dev, xdp->prog);
2816 	default:
2817 		return -EINVAL;
2818 	}
2819 }
2820 
2821 static const struct net_device_ops mlx4_netdev_ops = {
2822 	.ndo_open		= mlx4_en_open,
2823 	.ndo_stop		= mlx4_en_close,
2824 	.ndo_start_xmit		= mlx4_en_xmit,
2825 	.ndo_select_queue	= mlx4_en_select_queue,
2826 	.ndo_get_stats64	= mlx4_en_get_stats64,
2827 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2828 	.ndo_set_mac_address	= mlx4_en_set_mac,
2829 	.ndo_validate_addr	= eth_validate_addr,
2830 	.ndo_change_mtu		= mlx4_en_change_mtu,
2831 	.ndo_eth_ioctl		= mlx4_en_ioctl,
2832 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2833 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2834 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2835 	.ndo_set_features	= mlx4_en_set_features,
2836 	.ndo_fix_features	= mlx4_en_fix_features,
2837 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2838 #ifdef CONFIG_RFS_ACCEL
2839 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2840 #endif
2841 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2842 	.ndo_features_check	= mlx4_en_features_check,
2843 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2844 	.ndo_bpf		= mlx4_xdp,
2845 };
2846 
2847 static const struct net_device_ops mlx4_netdev_ops_master = {
2848 	.ndo_open		= mlx4_en_open,
2849 	.ndo_stop		= mlx4_en_close,
2850 	.ndo_start_xmit		= mlx4_en_xmit,
2851 	.ndo_select_queue	= mlx4_en_select_queue,
2852 	.ndo_get_stats64	= mlx4_en_get_stats64,
2853 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2854 	.ndo_set_mac_address	= mlx4_en_set_mac,
2855 	.ndo_validate_addr	= eth_validate_addr,
2856 	.ndo_change_mtu		= mlx4_en_change_mtu,
2857 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2858 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2859 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2860 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2861 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2862 	.ndo_set_vf_rate	= mlx4_en_set_vf_rate,
2863 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2864 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2865 	.ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2866 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2867 	.ndo_set_features	= mlx4_en_set_features,
2868 	.ndo_fix_features	= mlx4_en_fix_features,
2869 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2870 #ifdef CONFIG_RFS_ACCEL
2871 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2872 #endif
2873 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2874 	.ndo_features_check	= mlx4_en_features_check,
2875 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2876 	.ndo_bpf		= mlx4_xdp,
2877 };
2878 
2879 struct mlx4_en_bond {
2880 	struct work_struct work;
2881 	struct mlx4_en_priv *priv;
2882 	int is_bonded;
2883 	struct mlx4_port_map port_map;
2884 };
2885 
2886 static void mlx4_en_bond_work(struct work_struct *work)
2887 {
2888 	struct mlx4_en_bond *bond = container_of(work,
2889 						     struct mlx4_en_bond,
2890 						     work);
2891 	int err = 0;
2892 	struct mlx4_dev *dev = bond->priv->mdev->dev;
2893 
2894 	if (bond->is_bonded) {
2895 		if (!mlx4_is_bonded(dev)) {
2896 			err = mlx4_bond(dev);
2897 			if (err)
2898 				en_err(bond->priv, "Fail to bond device\n");
2899 		}
2900 		if (!err) {
2901 			err = mlx4_port_map_set(dev, &bond->port_map);
2902 			if (err)
2903 				en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2904 				       bond->port_map.port1,
2905 				       bond->port_map.port2,
2906 				       err);
2907 		}
2908 	} else if (mlx4_is_bonded(dev)) {
2909 		err = mlx4_unbond(dev);
2910 		if (err)
2911 			en_err(bond->priv, "Fail to unbond device\n");
2912 	}
2913 	dev_put(bond->priv->dev);
2914 	kfree(bond);
2915 }
2916 
2917 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2918 				   u8 v2p_p1, u8 v2p_p2)
2919 {
2920 	struct mlx4_en_bond *bond = NULL;
2921 
2922 	bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2923 	if (!bond)
2924 		return -ENOMEM;
2925 
2926 	INIT_WORK(&bond->work, mlx4_en_bond_work);
2927 	bond->priv = priv;
2928 	bond->is_bonded = is_bonded;
2929 	bond->port_map.port1 = v2p_p1;
2930 	bond->port_map.port2 = v2p_p2;
2931 	dev_hold(priv->dev);
2932 	queue_work(priv->mdev->workqueue, &bond->work);
2933 	return 0;
2934 }
2935 
2936 int mlx4_en_netdev_event(struct notifier_block *this,
2937 			 unsigned long event, void *ptr)
2938 {
2939 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2940 	u8 port = 0;
2941 	struct mlx4_en_dev *mdev;
2942 	struct mlx4_dev *dev;
2943 	int i, num_eth_ports = 0;
2944 	bool do_bond = true;
2945 	struct mlx4_en_priv *priv;
2946 	u8 v2p_port1 = 0;
2947 	u8 v2p_port2 = 0;
2948 
2949 	if (!net_eq(dev_net(ndev), &init_net))
2950 		return NOTIFY_DONE;
2951 
2952 	mdev = container_of(this, struct mlx4_en_dev, nb);
2953 	dev = mdev->dev;
2954 
2955 	/* Go into this mode only when two network devices set on two ports
2956 	 * of the same mlx4 device are slaves of the same bonding master
2957 	 */
2958 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2959 		++num_eth_ports;
2960 		if (!port && (mdev->pndev[i] == ndev))
2961 			port = i;
2962 		mdev->upper[i] = mdev->pndev[i] ?
2963 			netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2964 		/* condition not met: network device is a slave */
2965 		if (!mdev->upper[i])
2966 			do_bond = false;
2967 		if (num_eth_ports < 2)
2968 			continue;
2969 		/* condition not met: same master */
2970 		if (mdev->upper[i] != mdev->upper[i-1])
2971 			do_bond = false;
2972 	}
2973 	/* condition not met: 2 salves */
2974 	do_bond = (num_eth_ports ==  2) ? do_bond : false;
2975 
2976 	/* handle only events that come with enough info */
2977 	if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
2978 		return NOTIFY_DONE;
2979 
2980 	priv = netdev_priv(ndev);
2981 	if (do_bond) {
2982 		struct netdev_notifier_bonding_info *notifier_info = ptr;
2983 		struct netdev_bonding_info *bonding_info =
2984 			&notifier_info->bonding_info;
2985 
2986 		/* required mode 1, 2 or 4 */
2987 		if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
2988 		    (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
2989 		    (bonding_info->master.bond_mode != BOND_MODE_8023AD))
2990 			do_bond = false;
2991 
2992 		/* require exactly 2 slaves */
2993 		if (bonding_info->master.num_slaves != 2)
2994 			do_bond = false;
2995 
2996 		/* calc v2p */
2997 		if (do_bond) {
2998 			if (bonding_info->master.bond_mode ==
2999 			    BOND_MODE_ACTIVEBACKUP) {
3000 				/* in active-backup mode virtual ports are
3001 				 * mapped to the physical port of the active
3002 				 * slave */
3003 				if (bonding_info->slave.state ==
3004 				    BOND_STATE_BACKUP) {
3005 					if (port == 1) {
3006 						v2p_port1 = 2;
3007 						v2p_port2 = 2;
3008 					} else {
3009 						v2p_port1 = 1;
3010 						v2p_port2 = 1;
3011 					}
3012 				} else { /* BOND_STATE_ACTIVE */
3013 					if (port == 1) {
3014 						v2p_port1 = 1;
3015 						v2p_port2 = 1;
3016 					} else {
3017 						v2p_port1 = 2;
3018 						v2p_port2 = 2;
3019 					}
3020 				}
3021 			} else { /* Active-Active */
3022 				/* in active-active mode a virtual port is
3023 				 * mapped to the native physical port if and only
3024 				 * if the physical port is up */
3025 				__s8 link = bonding_info->slave.link;
3026 
3027 				if (port == 1)
3028 					v2p_port2 = 2;
3029 				else
3030 					v2p_port1 = 1;
3031 				if ((link == BOND_LINK_UP) ||
3032 				    (link == BOND_LINK_FAIL)) {
3033 					if (port == 1)
3034 						v2p_port1 = 1;
3035 					else
3036 						v2p_port2 = 2;
3037 				} else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3038 					if (port == 1)
3039 						v2p_port1 = 2;
3040 					else
3041 						v2p_port2 = 1;
3042 				}
3043 			}
3044 		}
3045 	}
3046 
3047 	mlx4_en_queue_bond_work(priv, do_bond,
3048 				v2p_port1, v2p_port2);
3049 
3050 	return NOTIFY_DONE;
3051 }
3052 
3053 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3054 				     struct mlx4_en_stats_bitmap *stats_bitmap,
3055 				     u8 rx_ppp, u8 rx_pause,
3056 				     u8 tx_ppp, u8 tx_pause)
3057 {
3058 	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3059 
3060 	if (!mlx4_is_slave(dev) &&
3061 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3062 		mutex_lock(&stats_bitmap->mutex);
3063 		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3064 
3065 		if (rx_ppp)
3066 			bitmap_set(stats_bitmap->bitmap, last_i,
3067 				   NUM_FLOW_PRIORITY_STATS_RX);
3068 		last_i += NUM_FLOW_PRIORITY_STATS_RX;
3069 
3070 		if (rx_pause && !(rx_ppp))
3071 			bitmap_set(stats_bitmap->bitmap, last_i,
3072 				   NUM_FLOW_STATS_RX);
3073 		last_i += NUM_FLOW_STATS_RX;
3074 
3075 		if (tx_ppp)
3076 			bitmap_set(stats_bitmap->bitmap, last_i,
3077 				   NUM_FLOW_PRIORITY_STATS_TX);
3078 		last_i += NUM_FLOW_PRIORITY_STATS_TX;
3079 
3080 		if (tx_pause && !(tx_ppp))
3081 			bitmap_set(stats_bitmap->bitmap, last_i,
3082 				   NUM_FLOW_STATS_TX);
3083 		last_i += NUM_FLOW_STATS_TX;
3084 
3085 		mutex_unlock(&stats_bitmap->mutex);
3086 	}
3087 }
3088 
3089 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3090 			      struct mlx4_en_stats_bitmap *stats_bitmap,
3091 			      u8 rx_ppp, u8 rx_pause,
3092 			      u8 tx_ppp, u8 tx_pause)
3093 {
3094 	int last_i = 0;
3095 
3096 	mutex_init(&stats_bitmap->mutex);
3097 	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3098 
3099 	if (mlx4_is_slave(dev)) {
3100 		bitmap_set(stats_bitmap->bitmap, last_i +
3101 					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3102 		bitmap_set(stats_bitmap->bitmap, last_i +
3103 					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3104 		bitmap_set(stats_bitmap->bitmap, last_i +
3105 					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3106 		bitmap_set(stats_bitmap->bitmap, last_i +
3107 					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3108 		bitmap_set(stats_bitmap->bitmap, last_i +
3109 					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3110 		bitmap_set(stats_bitmap->bitmap, last_i +
3111 					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3112 	} else {
3113 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3114 	}
3115 	last_i += NUM_MAIN_STATS;
3116 
3117 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3118 	last_i += NUM_PORT_STATS;
3119 
3120 	if (mlx4_is_master(dev))
3121 		bitmap_set(stats_bitmap->bitmap, last_i,
3122 			   NUM_PF_STATS);
3123 	last_i += NUM_PF_STATS;
3124 
3125 	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3126 					rx_ppp, rx_pause,
3127 					tx_ppp, tx_pause);
3128 	last_i += NUM_FLOW_STATS;
3129 
3130 	if (!mlx4_is_slave(dev))
3131 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3132 	last_i += NUM_PKT_STATS;
3133 
3134 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3135 	last_i += NUM_XDP_STATS;
3136 
3137 	if (!mlx4_is_slave(dev))
3138 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PHY_STATS);
3139 	last_i += NUM_PHY_STATS;
3140 }
3141 
3142 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3143 			struct mlx4_en_port_profile *prof)
3144 {
3145 	struct net_device *dev;
3146 	struct mlx4_en_priv *priv;
3147 	int i, t;
3148 	int err;
3149 
3150 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3151 				 MAX_TX_RINGS, MAX_RX_RINGS);
3152 	if (dev == NULL)
3153 		return -ENOMEM;
3154 
3155 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3156 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3157 
3158 	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3159 	dev->dev_port = port - 1;
3160 
3161 	/*
3162 	 * Initialize driver private data
3163 	 */
3164 
3165 	priv = netdev_priv(dev);
3166 	memset(priv, 0, sizeof(struct mlx4_en_priv));
3167 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3168 	spin_lock_init(&priv->stats_lock);
3169 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3170 	INIT_WORK(&priv->restart_task, mlx4_en_restart);
3171 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3172 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3173 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3174 #ifdef CONFIG_RFS_ACCEL
3175 	INIT_LIST_HEAD(&priv->filters);
3176 	spin_lock_init(&priv->filters_lock);
3177 #endif
3178 
3179 	priv->dev = dev;
3180 	priv->mdev = mdev;
3181 	priv->ddev = &mdev->pdev->dev;
3182 	priv->prof = prof;
3183 	priv->port = port;
3184 	priv->port_up = false;
3185 	priv->flags = prof->flags;
3186 	priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3187 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3188 			MLX4_WQE_CTRL_SOLICITED);
3189 	priv->num_tx_rings_p_up = mdev->profile.max_num_tx_rings_p_up;
3190 	priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3191 	netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3192 
3193 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3194 		priv->tx_ring_num[t] = prof->tx_ring_num[t];
3195 		if (!priv->tx_ring_num[t])
3196 			continue;
3197 
3198 		priv->tx_ring[t] = kcalloc(MAX_TX_RINGS,
3199 					   sizeof(struct mlx4_en_tx_ring *),
3200 					   GFP_KERNEL);
3201 		if (!priv->tx_ring[t]) {
3202 			err = -ENOMEM;
3203 			goto out;
3204 		}
3205 		priv->tx_cq[t] = kcalloc(MAX_TX_RINGS,
3206 					 sizeof(struct mlx4_en_cq *),
3207 					 GFP_KERNEL);
3208 		if (!priv->tx_cq[t]) {
3209 			err = -ENOMEM;
3210 			goto out;
3211 		}
3212 	}
3213 	priv->rx_ring_num = prof->rx_ring_num;
3214 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3215 	priv->cqe_size = mdev->dev->caps.cqe_size;
3216 	priv->mac_index = -1;
3217 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
3218 #ifdef CONFIG_MLX4_EN_DCB
3219 	if (!mlx4_is_slave(priv->mdev->dev)) {
3220 		u8 prio;
3221 
3222 		for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; ++prio) {
3223 			priv->ets.prio_tc[prio] = prio;
3224 			priv->ets.tc_tsa[prio]  = IEEE_8021QAZ_TSA_VENDOR;
3225 		}
3226 
3227 		priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3228 			DCB_CAP_DCBX_VER_IEEE;
3229 		priv->flags |= MLX4_EN_DCB_ENABLED;
3230 		priv->cee_config.pfc_state = false;
3231 
3232 		for (i = 0; i < MLX4_EN_NUM_UP_HIGH; i++)
3233 			priv->cee_config.dcb_pfc[i] = pfc_disabled;
3234 
3235 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3236 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3237 		} else {
3238 			en_info(priv, "enabling only PFC DCB ops\n");
3239 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3240 		}
3241 	}
3242 #endif
3243 
3244 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3245 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
3246 
3247 	/* Query for default mac and max mtu */
3248 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3249 
3250 	if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3251 	    MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3252 		priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3253 
3254 	/* Set default MAC */
3255 	dev->addr_len = ETH_ALEN;
3256 	mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3257 	if (!is_valid_ether_addr(dev->dev_addr)) {
3258 		en_err(priv, "Port: %d, invalid mac burned: %pM, quitting\n",
3259 		       priv->port, dev->dev_addr);
3260 		err = -EINVAL;
3261 		goto out;
3262 	} else if (mlx4_is_slave(priv->mdev->dev) &&
3263 		   (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3264 		/* Random MAC was assigned in mlx4_slave_cap
3265 		 * in mlx4_core module
3266 		 */
3267 		dev->addr_assign_type |= NET_ADDR_RANDOM;
3268 		en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3269 	}
3270 
3271 	memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3272 
3273 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3274 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3275 	err = mlx4_en_alloc_resources(priv);
3276 	if (err)
3277 		goto out;
3278 
3279 	/* Initialize time stamping config */
3280 	priv->hwtstamp_config.flags = 0;
3281 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3282 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3283 
3284 	/* Allocate page for receive rings */
3285 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3286 				MLX4_EN_PAGE_SIZE);
3287 	if (err) {
3288 		en_err(priv, "Failed to allocate page for rx qps\n");
3289 		goto out;
3290 	}
3291 	priv->allocated = 1;
3292 
3293 	/*
3294 	 * Initialize netdev entry points
3295 	 */
3296 	if (mlx4_is_master(priv->mdev->dev))
3297 		dev->netdev_ops = &mlx4_netdev_ops_master;
3298 	else
3299 		dev->netdev_ops = &mlx4_netdev_ops;
3300 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3301 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3302 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3303 
3304 	dev->ethtool_ops = &mlx4_en_ethtool_ops;
3305 
3306 	/*
3307 	 * Set driver features
3308 	 */
3309 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3310 	if (mdev->LSO_support)
3311 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3312 
3313 	if (mdev->dev->caps.tunnel_offload_mode ==
3314 	    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3315 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3316 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3317 				    NETIF_F_GSO_PARTIAL;
3318 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3319 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3320 				    NETIF_F_GSO_PARTIAL;
3321 		dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3322 		dev->hw_enc_features = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3323 				       NETIF_F_RXCSUM |
3324 				       NETIF_F_TSO | NETIF_F_TSO6 |
3325 				       NETIF_F_GSO_UDP_TUNNEL |
3326 				       NETIF_F_GSO_UDP_TUNNEL_CSUM |
3327 				       NETIF_F_GSO_PARTIAL;
3328 
3329 		dev->udp_tunnel_nic_info = &mlx4_udp_tunnels;
3330 	}
3331 
3332 	dev->vlan_features = dev->hw_features;
3333 
3334 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3335 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3336 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3337 			NETIF_F_HW_VLAN_CTAG_FILTER;
3338 	dev->hw_features |= NETIF_F_LOOPBACK |
3339 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3340 
3341 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3342 		dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3343 			NETIF_F_HW_VLAN_STAG_FILTER;
3344 		dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3345 	}
3346 
3347 	if (mlx4_is_slave(mdev->dev)) {
3348 		bool vlan_offload_disabled;
3349 		int phv;
3350 
3351 		err = get_phv_bit(mdev->dev, port, &phv);
3352 		if (!err && phv) {
3353 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3354 			priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3355 		}
3356 		err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3357 							&vlan_offload_disabled);
3358 		if (!err && vlan_offload_disabled) {
3359 			dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3360 					      NETIF_F_HW_VLAN_CTAG_RX |
3361 					      NETIF_F_HW_VLAN_STAG_TX |
3362 					      NETIF_F_HW_VLAN_STAG_RX);
3363 			dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3364 					   NETIF_F_HW_VLAN_CTAG_RX |
3365 					   NETIF_F_HW_VLAN_STAG_TX |
3366 					   NETIF_F_HW_VLAN_STAG_RX);
3367 		}
3368 	} else {
3369 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3370 		    !(mdev->dev->caps.flags2 &
3371 		      MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3372 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3373 	}
3374 
3375 	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3376 		dev->hw_features |= NETIF_F_RXFCS;
3377 
3378 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3379 		dev->hw_features |= NETIF_F_RXALL;
3380 
3381 	if (mdev->dev->caps.steering_mode ==
3382 	    MLX4_STEERING_MODE_DEVICE_MANAGED &&
3383 	    mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3384 		dev->hw_features |= NETIF_F_NTUPLE;
3385 
3386 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3387 		dev->priv_flags |= IFF_UNICAST_FLT;
3388 
3389 	/* Setting a default hash function value */
3390 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3391 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3392 	} else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3393 		priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3394 	} else {
3395 		en_warn(priv,
3396 			"No RSS hash capabilities exposed, using Toeplitz\n");
3397 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3398 	}
3399 
3400 	/* MTU range: 68 - hw-specific max */
3401 	dev->min_mtu = ETH_MIN_MTU;
3402 	dev->max_mtu = priv->max_mtu;
3403 
3404 	mdev->pndev[port] = dev;
3405 	mdev->upper[port] = NULL;
3406 
3407 	netif_carrier_off(dev);
3408 	mlx4_en_set_default_moderation(priv);
3409 
3410 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3411 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3412 
3413 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3414 
3415 	/* Configure port */
3416 	mlx4_en_calc_rx_buf(dev);
3417 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3418 				    priv->rx_skb_size + ETH_FCS_LEN,
3419 				    prof->tx_pause, prof->tx_ppp,
3420 				    prof->rx_pause, prof->rx_ppp);
3421 	if (err) {
3422 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3423 		       priv->port, err);
3424 		goto out;
3425 	}
3426 
3427 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3428 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3429 		if (err) {
3430 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3431 			       err);
3432 			goto out;
3433 		}
3434 	}
3435 
3436 	/* Init port */
3437 	en_warn(priv, "Initializing port\n");
3438 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
3439 	if (err) {
3440 		en_err(priv, "Failed Initializing port\n");
3441 		goto out;
3442 	}
3443 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3444 
3445 	/* Initialize time stamp mechanism */
3446 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3447 		mlx4_en_init_timestamp(mdev);
3448 
3449 	queue_delayed_work(mdev->workqueue, &priv->service_task,
3450 			   SERVICE_TASK_DELAY);
3451 
3452 	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3453 				 mdev->profile.prof[priv->port].rx_ppp,
3454 				 mdev->profile.prof[priv->port].rx_pause,
3455 				 mdev->profile.prof[priv->port].tx_ppp,
3456 				 mdev->profile.prof[priv->port].tx_pause);
3457 
3458 	err = register_netdev(dev);
3459 	if (err) {
3460 		en_err(priv, "Netdev registration failed for port %d\n", port);
3461 		goto out;
3462 	}
3463 
3464 	priv->registered = 1;
3465 	devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3466 				  dev);
3467 
3468 	return 0;
3469 
3470 out:
3471 	mlx4_en_destroy_netdev(dev);
3472 	return err;
3473 }
3474 
3475 int mlx4_en_reset_config(struct net_device *dev,
3476 			 struct hwtstamp_config ts_config,
3477 			 netdev_features_t features)
3478 {
3479 	struct mlx4_en_priv *priv = netdev_priv(dev);
3480 	struct mlx4_en_dev *mdev = priv->mdev;
3481 	struct mlx4_en_port_profile new_prof;
3482 	struct mlx4_en_priv *tmp;
3483 	int port_up = 0;
3484 	int err = 0;
3485 
3486 	if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3487 	    priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3488 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3489 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3490 		return 0; /* Nothing to change */
3491 
3492 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3493 	    (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3494 	    (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3495 		en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3496 		return -EINVAL;
3497 	}
3498 
3499 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3500 	if (!tmp)
3501 		return -ENOMEM;
3502 
3503 	mutex_lock(&mdev->state_lock);
3504 
3505 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3506 	memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3507 
3508 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3509 	if (err)
3510 		goto out;
3511 
3512 	if (priv->port_up) {
3513 		port_up = 1;
3514 		mlx4_en_stop_port(dev, 1);
3515 	}
3516 
3517 	mlx4_en_safe_replace_resources(priv, tmp);
3518 
3519 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3520 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3521 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3522 		else
3523 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3524 	} else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3525 		/* RX time-stamping is OFF, update the RX vlan offload
3526 		 * to the latest wanted state
3527 		 */
3528 		if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3529 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3530 		else
3531 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3532 	}
3533 
3534 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3535 		if (features & NETIF_F_RXFCS)
3536 			dev->features |= NETIF_F_RXFCS;
3537 		else
3538 			dev->features &= ~NETIF_F_RXFCS;
3539 	}
3540 
3541 	/* RX vlan offload and RX time-stamping can't co-exist !
3542 	 * Regardless of the caller's choice,
3543 	 * Turn Off RX vlan offload in case of time-stamping is ON
3544 	 */
3545 	if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3546 		if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3547 			en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3548 		dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3549 	}
3550 
3551 	if (port_up) {
3552 		err = mlx4_en_start_port(dev);
3553 		if (err)
3554 			en_err(priv, "Failed starting port\n");
3555 	}
3556 
3557 	if (!err)
3558 		err = mlx4_en_moderation_update(priv);
3559 out:
3560 	mutex_unlock(&mdev->state_lock);
3561 	kfree(tmp);
3562 	if (!err)
3563 		netdev_features_change(dev);
3564 	return err;
3565 }
3566