xref: /linux/drivers/net/ethernet/mellanox/mlx4/en_netdev.c (revision c1aac62f36c1e37ee81c9e09ee9ee733eef05dcb)
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/busy_poll.h>
43 #include <net/vxlan.h>
44 #include <net/devlink.h>
45 
46 #include <linux/mlx4/driver.h>
47 #include <linux/mlx4/device.h>
48 #include <linux/mlx4/cmd.h>
49 #include <linux/mlx4/cq.h>
50 
51 #include "mlx4_en.h"
52 #include "en_port.h"
53 
54 #define MLX4_EN_MAX_XDP_MTU ((int)(PAGE_SIZE - ETH_HLEN - (2 * VLAN_HLEN) - \
55 				   XDP_PACKET_HEADROOM))
56 
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)
64 		return -EINVAL;
65 
66 	netdev_set_num_tc(dev, up);
67 
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 static int __mlx4_en_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
90 			      struct tc_to_netdev *tc)
91 {
92 	if (tc->type != TC_SETUP_MQPRIO)
93 		return -EINVAL;
94 
95 	return mlx4_en_setup_tc(dev, tc->tc);
96 }
97 
98 #ifdef CONFIG_RFS_ACCEL
99 
100 struct mlx4_en_filter {
101 	struct list_head next;
102 	struct work_struct work;
103 
104 	u8     ip_proto;
105 	__be32 src_ip;
106 	__be32 dst_ip;
107 	__be16 src_port;
108 	__be16 dst_port;
109 
110 	int rxq_index;
111 	struct mlx4_en_priv *priv;
112 	u32 flow_id;			/* RFS infrastructure id */
113 	int id;				/* mlx4_en driver id */
114 	u64 reg_id;			/* Flow steering API id */
115 	u8 activated;			/* Used to prevent expiry before filter
116 					 * is attached
117 					 */
118 	struct hlist_node filter_chain;
119 };
120 
121 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
122 
123 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
124 {
125 	switch (ip_proto) {
126 	case IPPROTO_UDP:
127 		return MLX4_NET_TRANS_RULE_ID_UDP;
128 	case IPPROTO_TCP:
129 		return MLX4_NET_TRANS_RULE_ID_TCP;
130 	default:
131 		return MLX4_NET_TRANS_RULE_NUM;
132 	}
133 };
134 
135 /* Must not acquire state_lock, as its corresponding work_sync
136  * is done under it.
137  */
138 static void mlx4_en_filter_work(struct work_struct *work)
139 {
140 	struct mlx4_en_filter *filter = container_of(work,
141 						     struct mlx4_en_filter,
142 						     work);
143 	struct mlx4_en_priv *priv = filter->priv;
144 	struct mlx4_spec_list spec_tcp_udp = {
145 		.id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
146 		{
147 			.tcp_udp = {
148 				.dst_port = filter->dst_port,
149 				.dst_port_msk = (__force __be16)-1,
150 				.src_port = filter->src_port,
151 				.src_port_msk = (__force __be16)-1,
152 			},
153 		},
154 	};
155 	struct mlx4_spec_list spec_ip = {
156 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
157 		{
158 			.ipv4 = {
159 				.dst_ip = filter->dst_ip,
160 				.dst_ip_msk = (__force __be32)-1,
161 				.src_ip = filter->src_ip,
162 				.src_ip_msk = (__force __be32)-1,
163 			},
164 		},
165 	};
166 	struct mlx4_spec_list spec_eth = {
167 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
168 	};
169 	struct mlx4_net_trans_rule rule = {
170 		.list = LIST_HEAD_INIT(rule.list),
171 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
172 		.exclusive = 1,
173 		.allow_loopback = 1,
174 		.promisc_mode = MLX4_FS_REGULAR,
175 		.port = priv->port,
176 		.priority = MLX4_DOMAIN_RFS,
177 	};
178 	int rc;
179 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
180 
181 	if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
182 		en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
183 			filter->ip_proto);
184 		goto ignore;
185 	}
186 	list_add_tail(&spec_eth.list, &rule.list);
187 	list_add_tail(&spec_ip.list, &rule.list);
188 	list_add_tail(&spec_tcp_udp.list, &rule.list);
189 
190 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
191 	memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
192 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
193 
194 	filter->activated = 0;
195 
196 	if (filter->reg_id) {
197 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
198 		if (rc && rc != -ENOENT)
199 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
200 	}
201 
202 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
203 	if (rc)
204 		en_err(priv, "Error attaching flow. err = %d\n", rc);
205 
206 ignore:
207 	mlx4_en_filter_rfs_expire(priv);
208 
209 	filter->activated = 1;
210 }
211 
212 static inline struct hlist_head *
213 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
214 		   __be16 src_port, __be16 dst_port)
215 {
216 	unsigned long l;
217 	int bucket_idx;
218 
219 	l = (__force unsigned long)src_port |
220 	    ((__force unsigned long)dst_port << 2);
221 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
222 
223 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
224 
225 	return &priv->filter_hash[bucket_idx];
226 }
227 
228 static struct mlx4_en_filter *
229 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
230 		     __be32 dst_ip, u8 ip_proto, __be16 src_port,
231 		     __be16 dst_port, u32 flow_id)
232 {
233 	struct mlx4_en_filter *filter = NULL;
234 
235 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
236 	if (!filter)
237 		return NULL;
238 
239 	filter->priv = priv;
240 	filter->rxq_index = rxq_index;
241 	INIT_WORK(&filter->work, mlx4_en_filter_work);
242 
243 	filter->src_ip = src_ip;
244 	filter->dst_ip = dst_ip;
245 	filter->ip_proto = ip_proto;
246 	filter->src_port = src_port;
247 	filter->dst_port = dst_port;
248 
249 	filter->flow_id = flow_id;
250 
251 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
252 
253 	list_add_tail(&filter->next, &priv->filters);
254 	hlist_add_head(&filter->filter_chain,
255 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
256 					  dst_port));
257 
258 	return filter;
259 }
260 
261 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
262 {
263 	struct mlx4_en_priv *priv = filter->priv;
264 	int rc;
265 
266 	list_del(&filter->next);
267 
268 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
269 	if (rc && rc != -ENOENT)
270 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
271 
272 	kfree(filter);
273 }
274 
275 static inline struct mlx4_en_filter *
276 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
277 		    u8 ip_proto, __be16 src_port, __be16 dst_port)
278 {
279 	struct mlx4_en_filter *filter;
280 	struct mlx4_en_filter *ret = NULL;
281 
282 	hlist_for_each_entry(filter,
283 			     filter_hash_bucket(priv, src_ip, dst_ip,
284 						src_port, dst_port),
285 			     filter_chain) {
286 		if (filter->src_ip == src_ip &&
287 		    filter->dst_ip == dst_ip &&
288 		    filter->ip_proto == ip_proto &&
289 		    filter->src_port == src_port &&
290 		    filter->dst_port == dst_port) {
291 			ret = filter;
292 			break;
293 		}
294 	}
295 
296 	return ret;
297 }
298 
299 static int
300 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
301 		   u16 rxq_index, u32 flow_id)
302 {
303 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
304 	struct mlx4_en_filter *filter;
305 	const struct iphdr *ip;
306 	const __be16 *ports;
307 	u8 ip_proto;
308 	__be32 src_ip;
309 	__be32 dst_ip;
310 	__be16 src_port;
311 	__be16 dst_port;
312 	int nhoff = skb_network_offset(skb);
313 	int ret = 0;
314 
315 	if (skb->protocol != htons(ETH_P_IP))
316 		return -EPROTONOSUPPORT;
317 
318 	ip = (const struct iphdr *)(skb->data + nhoff);
319 	if (ip_is_fragment(ip))
320 		return -EPROTONOSUPPORT;
321 
322 	if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
323 		return -EPROTONOSUPPORT;
324 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
325 
326 	ip_proto = ip->protocol;
327 	src_ip = ip->saddr;
328 	dst_ip = ip->daddr;
329 	src_port = ports[0];
330 	dst_port = ports[1];
331 
332 	spin_lock_bh(&priv->filters_lock);
333 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
334 				     src_port, dst_port);
335 	if (filter) {
336 		if (filter->rxq_index == rxq_index)
337 			goto out;
338 
339 		filter->rxq_index = rxq_index;
340 	} else {
341 		filter = mlx4_en_filter_alloc(priv, rxq_index,
342 					      src_ip, dst_ip, ip_proto,
343 					      src_port, dst_port, flow_id);
344 		if (!filter) {
345 			ret = -ENOMEM;
346 			goto err;
347 		}
348 	}
349 
350 	queue_work(priv->mdev->workqueue, &filter->work);
351 
352 out:
353 	ret = filter->id;
354 err:
355 	spin_unlock_bh(&priv->filters_lock);
356 
357 	return ret;
358 }
359 
360 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
361 {
362 	struct mlx4_en_filter *filter, *tmp;
363 	LIST_HEAD(del_list);
364 
365 	spin_lock_bh(&priv->filters_lock);
366 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
367 		list_move(&filter->next, &del_list);
368 		hlist_del(&filter->filter_chain);
369 	}
370 	spin_unlock_bh(&priv->filters_lock);
371 
372 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
373 		cancel_work_sync(&filter->work);
374 		mlx4_en_filter_free(filter);
375 	}
376 }
377 
378 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
379 {
380 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
381 	LIST_HEAD(del_list);
382 	int i = 0;
383 
384 	spin_lock_bh(&priv->filters_lock);
385 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
386 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
387 			break;
388 
389 		if (filter->activated &&
390 		    !work_pending(&filter->work) &&
391 		    rps_may_expire_flow(priv->dev,
392 					filter->rxq_index, filter->flow_id,
393 					filter->id)) {
394 			list_move(&filter->next, &del_list);
395 			hlist_del(&filter->filter_chain);
396 		} else
397 			last_filter = filter;
398 
399 		i++;
400 	}
401 
402 	if (last_filter && (&last_filter->next != priv->filters.next))
403 		list_move(&priv->filters, &last_filter->next);
404 
405 	spin_unlock_bh(&priv->filters_lock);
406 
407 	list_for_each_entry_safe(filter, tmp, &del_list, next)
408 		mlx4_en_filter_free(filter);
409 }
410 #endif
411 
412 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
413 				   __be16 proto, u16 vid)
414 {
415 	struct mlx4_en_priv *priv = netdev_priv(dev);
416 	struct mlx4_en_dev *mdev = priv->mdev;
417 	int err;
418 	int idx;
419 
420 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
421 
422 	set_bit(vid, priv->active_vlans);
423 
424 	/* Add VID to port VLAN filter */
425 	mutex_lock(&mdev->state_lock);
426 	if (mdev->device_up && priv->port_up) {
427 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
428 		if (err) {
429 			en_err(priv, "Failed configuring VLAN filter\n");
430 			goto out;
431 		}
432 	}
433 	err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
434 	if (err)
435 		en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
436 
437 out:
438 	mutex_unlock(&mdev->state_lock);
439 	return err;
440 }
441 
442 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
443 				    __be16 proto, u16 vid)
444 {
445 	struct mlx4_en_priv *priv = netdev_priv(dev);
446 	struct mlx4_en_dev *mdev = priv->mdev;
447 	int err = 0;
448 
449 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
450 
451 	clear_bit(vid, priv->active_vlans);
452 
453 	/* Remove VID from port VLAN filter */
454 	mutex_lock(&mdev->state_lock);
455 	mlx4_unregister_vlan(mdev->dev, priv->port, vid);
456 
457 	if (mdev->device_up && priv->port_up) {
458 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
459 		if (err)
460 			en_err(priv, "Failed configuring VLAN filter\n");
461 	}
462 	mutex_unlock(&mdev->state_lock);
463 
464 	return err;
465 }
466 
467 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
468 {
469 	int i;
470 	for (i = ETH_ALEN - 1; i >= 0; --i) {
471 		dst_mac[i] = src_mac & 0xff;
472 		src_mac >>= 8;
473 	}
474 	memset(&dst_mac[ETH_ALEN], 0, 2);
475 }
476 
477 
478 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
479 				    int qpn, u64 *reg_id)
480 {
481 	int err;
482 
483 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
484 	    priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
485 		return 0; /* do nothing */
486 
487 	err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
488 				    MLX4_DOMAIN_NIC, reg_id);
489 	if (err) {
490 		en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
491 		return err;
492 	}
493 	en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
494 	return 0;
495 }
496 
497 
498 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
499 				unsigned char *mac, int *qpn, u64 *reg_id)
500 {
501 	struct mlx4_en_dev *mdev = priv->mdev;
502 	struct mlx4_dev *dev = mdev->dev;
503 	int err;
504 
505 	switch (dev->caps.steering_mode) {
506 	case MLX4_STEERING_MODE_B0: {
507 		struct mlx4_qp qp;
508 		u8 gid[16] = {0};
509 
510 		qp.qpn = *qpn;
511 		memcpy(&gid[10], mac, ETH_ALEN);
512 		gid[5] = priv->port;
513 
514 		err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
515 		break;
516 	}
517 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
518 		struct mlx4_spec_list spec_eth = { {NULL} };
519 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
520 
521 		struct mlx4_net_trans_rule rule = {
522 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
523 			.exclusive = 0,
524 			.allow_loopback = 1,
525 			.promisc_mode = MLX4_FS_REGULAR,
526 			.priority = MLX4_DOMAIN_NIC,
527 		};
528 
529 		rule.port = priv->port;
530 		rule.qpn = *qpn;
531 		INIT_LIST_HEAD(&rule.list);
532 
533 		spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
534 		memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
535 		memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
536 		list_add_tail(&spec_eth.list, &rule.list);
537 
538 		err = mlx4_flow_attach(dev, &rule, reg_id);
539 		break;
540 	}
541 	default:
542 		return -EINVAL;
543 	}
544 	if (err)
545 		en_warn(priv, "Failed Attaching Unicast\n");
546 
547 	return err;
548 }
549 
550 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
551 				     unsigned char *mac, int qpn, u64 reg_id)
552 {
553 	struct mlx4_en_dev *mdev = priv->mdev;
554 	struct mlx4_dev *dev = mdev->dev;
555 
556 	switch (dev->caps.steering_mode) {
557 	case MLX4_STEERING_MODE_B0: {
558 		struct mlx4_qp qp;
559 		u8 gid[16] = {0};
560 
561 		qp.qpn = qpn;
562 		memcpy(&gid[10], mac, ETH_ALEN);
563 		gid[5] = priv->port;
564 
565 		mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
566 		break;
567 	}
568 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
569 		mlx4_flow_detach(dev, reg_id);
570 		break;
571 	}
572 	default:
573 		en_err(priv, "Invalid steering mode.\n");
574 	}
575 }
576 
577 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
578 {
579 	struct mlx4_en_dev *mdev = priv->mdev;
580 	struct mlx4_dev *dev = mdev->dev;
581 	int index = 0;
582 	int err = 0;
583 	int *qpn = &priv->base_qpn;
584 	u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
585 
586 	en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
587 	       priv->dev->dev_addr);
588 	index = mlx4_register_mac(dev, priv->port, mac);
589 	if (index < 0) {
590 		err = index;
591 		en_err(priv, "Failed adding MAC: %pM\n",
592 		       priv->dev->dev_addr);
593 		return err;
594 	}
595 
596 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
597 		int base_qpn = mlx4_get_base_qpn(dev, priv->port);
598 		*qpn = base_qpn + index;
599 		return 0;
600 	}
601 
602 	err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP);
603 	en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
604 	if (err) {
605 		en_err(priv, "Failed to reserve qp for mac registration\n");
606 		mlx4_unregister_mac(dev, priv->port, mac);
607 		return err;
608 	}
609 
610 	return 0;
611 }
612 
613 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
614 {
615 	struct mlx4_en_dev *mdev = priv->mdev;
616 	struct mlx4_dev *dev = mdev->dev;
617 	int qpn = priv->base_qpn;
618 
619 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
620 		u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
621 		en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
622 		       priv->dev->dev_addr);
623 		mlx4_unregister_mac(dev, priv->port, mac);
624 	} else {
625 		en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
626 		       priv->port, qpn);
627 		mlx4_qp_release_range(dev, qpn, 1);
628 		priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
629 	}
630 }
631 
632 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
633 			       unsigned char *new_mac, unsigned char *prev_mac)
634 {
635 	struct mlx4_en_dev *mdev = priv->mdev;
636 	struct mlx4_dev *dev = mdev->dev;
637 	int err = 0;
638 	u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
639 
640 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
641 		struct hlist_head *bucket;
642 		unsigned int mac_hash;
643 		struct mlx4_mac_entry *entry;
644 		struct hlist_node *tmp;
645 		u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
646 
647 		bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
648 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
649 			if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
650 				mlx4_en_uc_steer_release(priv, entry->mac,
651 							 qpn, entry->reg_id);
652 				mlx4_unregister_mac(dev, priv->port,
653 						    prev_mac_u64);
654 				hlist_del_rcu(&entry->hlist);
655 				synchronize_rcu();
656 				memcpy(entry->mac, new_mac, ETH_ALEN);
657 				entry->reg_id = 0;
658 				mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
659 				hlist_add_head_rcu(&entry->hlist,
660 						   &priv->mac_hash[mac_hash]);
661 				mlx4_register_mac(dev, priv->port, new_mac_u64);
662 				err = mlx4_en_uc_steer_add(priv, new_mac,
663 							   &qpn,
664 							   &entry->reg_id);
665 				if (err)
666 					return err;
667 				if (priv->tunnel_reg_id) {
668 					mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
669 					priv->tunnel_reg_id = 0;
670 				}
671 				err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
672 							       &priv->tunnel_reg_id);
673 				return err;
674 			}
675 		}
676 		return -EINVAL;
677 	}
678 
679 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
680 }
681 
682 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
683 			      unsigned char new_mac[ETH_ALEN + 2])
684 {
685 	int err = 0;
686 
687 	if (priv->port_up) {
688 		/* Remove old MAC and insert the new one */
689 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
690 					  new_mac, priv->current_mac);
691 		if (err)
692 			en_err(priv, "Failed changing HW MAC address\n");
693 	} else
694 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
695 
696 	if (!err)
697 		memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
698 
699 	return err;
700 }
701 
702 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
703 {
704 	struct mlx4_en_priv *priv = netdev_priv(dev);
705 	struct mlx4_en_dev *mdev = priv->mdev;
706 	struct sockaddr *saddr = addr;
707 	unsigned char new_mac[ETH_ALEN + 2];
708 	int err;
709 
710 	if (!is_valid_ether_addr(saddr->sa_data))
711 		return -EADDRNOTAVAIL;
712 
713 	mutex_lock(&mdev->state_lock);
714 	memcpy(new_mac, saddr->sa_data, ETH_ALEN);
715 	err = mlx4_en_do_set_mac(priv, new_mac);
716 	if (!err)
717 		memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
718 	mutex_unlock(&mdev->state_lock);
719 
720 	return err;
721 }
722 
723 static void mlx4_en_clear_list(struct net_device *dev)
724 {
725 	struct mlx4_en_priv *priv = netdev_priv(dev);
726 	struct mlx4_en_mc_list *tmp, *mc_to_del;
727 
728 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
729 		list_del(&mc_to_del->list);
730 		kfree(mc_to_del);
731 	}
732 }
733 
734 static void mlx4_en_cache_mclist(struct net_device *dev)
735 {
736 	struct mlx4_en_priv *priv = netdev_priv(dev);
737 	struct netdev_hw_addr *ha;
738 	struct mlx4_en_mc_list *tmp;
739 
740 	mlx4_en_clear_list(dev);
741 	netdev_for_each_mc_addr(ha, dev) {
742 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
743 		if (!tmp) {
744 			mlx4_en_clear_list(dev);
745 			return;
746 		}
747 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
748 		list_add_tail(&tmp->list, &priv->mc_list);
749 	}
750 }
751 
752 static void update_mclist_flags(struct mlx4_en_priv *priv,
753 				struct list_head *dst,
754 				struct list_head *src)
755 {
756 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
757 	bool found;
758 
759 	/* Find all the entries that should be removed from dst,
760 	 * These are the entries that are not found in src
761 	 */
762 	list_for_each_entry(dst_tmp, dst, list) {
763 		found = false;
764 		list_for_each_entry(src_tmp, src, list) {
765 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
766 				found = true;
767 				break;
768 			}
769 		}
770 		if (!found)
771 			dst_tmp->action = MCLIST_REM;
772 	}
773 
774 	/* Add entries that exist in src but not in dst
775 	 * mark them as need to add
776 	 */
777 	list_for_each_entry(src_tmp, src, list) {
778 		found = false;
779 		list_for_each_entry(dst_tmp, dst, list) {
780 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
781 				dst_tmp->action = MCLIST_NONE;
782 				found = true;
783 				break;
784 			}
785 		}
786 		if (!found) {
787 			new_mc = kmemdup(src_tmp,
788 					 sizeof(struct mlx4_en_mc_list),
789 					 GFP_KERNEL);
790 			if (!new_mc)
791 				return;
792 
793 			new_mc->action = MCLIST_ADD;
794 			list_add_tail(&new_mc->list, dst);
795 		}
796 	}
797 }
798 
799 static void mlx4_en_set_rx_mode(struct net_device *dev)
800 {
801 	struct mlx4_en_priv *priv = netdev_priv(dev);
802 
803 	if (!priv->port_up)
804 		return;
805 
806 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
807 }
808 
809 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
810 				     struct mlx4_en_dev *mdev)
811 {
812 	int err = 0;
813 
814 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
815 		if (netif_msg_rx_status(priv))
816 			en_warn(priv, "Entering promiscuous mode\n");
817 		priv->flags |= MLX4_EN_FLAG_PROMISC;
818 
819 		/* Enable promiscouos mode */
820 		switch (mdev->dev->caps.steering_mode) {
821 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
822 			err = mlx4_flow_steer_promisc_add(mdev->dev,
823 							  priv->port,
824 							  priv->base_qpn,
825 							  MLX4_FS_ALL_DEFAULT);
826 			if (err)
827 				en_err(priv, "Failed enabling promiscuous mode\n");
828 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
829 			break;
830 
831 		case MLX4_STEERING_MODE_B0:
832 			err = mlx4_unicast_promisc_add(mdev->dev,
833 						       priv->base_qpn,
834 						       priv->port);
835 			if (err)
836 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
837 
838 			/* Add the default qp number as multicast
839 			 * promisc
840 			 */
841 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
842 				err = mlx4_multicast_promisc_add(mdev->dev,
843 								 priv->base_qpn,
844 								 priv->port);
845 				if (err)
846 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
847 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
848 			}
849 			break;
850 
851 		case MLX4_STEERING_MODE_A0:
852 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
853 						     priv->port,
854 						     priv->base_qpn,
855 						     1);
856 			if (err)
857 				en_err(priv, "Failed enabling promiscuous mode\n");
858 			break;
859 		}
860 
861 		/* Disable port multicast filter (unconditionally) */
862 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
863 					  0, MLX4_MCAST_DISABLE);
864 		if (err)
865 			en_err(priv, "Failed disabling multicast filter\n");
866 	}
867 }
868 
869 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
870 				       struct mlx4_en_dev *mdev)
871 {
872 	int err = 0;
873 
874 	if (netif_msg_rx_status(priv))
875 		en_warn(priv, "Leaving promiscuous mode\n");
876 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
877 
878 	/* Disable promiscouos mode */
879 	switch (mdev->dev->caps.steering_mode) {
880 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
881 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
882 						     priv->port,
883 						     MLX4_FS_ALL_DEFAULT);
884 		if (err)
885 			en_err(priv, "Failed disabling promiscuous mode\n");
886 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
887 		break;
888 
889 	case MLX4_STEERING_MODE_B0:
890 		err = mlx4_unicast_promisc_remove(mdev->dev,
891 						  priv->base_qpn,
892 						  priv->port);
893 		if (err)
894 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
895 		/* Disable Multicast promisc */
896 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
897 			err = mlx4_multicast_promisc_remove(mdev->dev,
898 							    priv->base_qpn,
899 							    priv->port);
900 			if (err)
901 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
902 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
903 		}
904 		break;
905 
906 	case MLX4_STEERING_MODE_A0:
907 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
908 					     priv->port,
909 					     priv->base_qpn, 0);
910 		if (err)
911 			en_err(priv, "Failed disabling promiscuous mode\n");
912 		break;
913 	}
914 }
915 
916 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
917 				 struct net_device *dev,
918 				 struct mlx4_en_dev *mdev)
919 {
920 	struct mlx4_en_mc_list *mclist, *tmp;
921 	u64 mcast_addr = 0;
922 	u8 mc_list[16] = {0};
923 	int err = 0;
924 
925 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
926 	if (dev->flags & IFF_ALLMULTI) {
927 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
928 					  0, MLX4_MCAST_DISABLE);
929 		if (err)
930 			en_err(priv, "Failed disabling multicast filter\n");
931 
932 		/* Add the default qp number as multicast promisc */
933 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
934 			switch (mdev->dev->caps.steering_mode) {
935 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
936 				err = mlx4_flow_steer_promisc_add(mdev->dev,
937 								  priv->port,
938 								  priv->base_qpn,
939 								  MLX4_FS_MC_DEFAULT);
940 				break;
941 
942 			case MLX4_STEERING_MODE_B0:
943 				err = mlx4_multicast_promisc_add(mdev->dev,
944 								 priv->base_qpn,
945 								 priv->port);
946 				break;
947 
948 			case MLX4_STEERING_MODE_A0:
949 				break;
950 			}
951 			if (err)
952 				en_err(priv, "Failed entering multicast promisc mode\n");
953 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
954 		}
955 	} else {
956 		/* Disable Multicast promisc */
957 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
958 			switch (mdev->dev->caps.steering_mode) {
959 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
960 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
961 								     priv->port,
962 								     MLX4_FS_MC_DEFAULT);
963 				break;
964 
965 			case MLX4_STEERING_MODE_B0:
966 				err = mlx4_multicast_promisc_remove(mdev->dev,
967 								    priv->base_qpn,
968 								    priv->port);
969 				break;
970 
971 			case MLX4_STEERING_MODE_A0:
972 				break;
973 			}
974 			if (err)
975 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
976 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
977 		}
978 
979 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
980 					  0, MLX4_MCAST_DISABLE);
981 		if (err)
982 			en_err(priv, "Failed disabling multicast filter\n");
983 
984 		/* Flush mcast filter and init it with broadcast address */
985 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
986 				    1, MLX4_MCAST_CONFIG);
987 
988 		/* Update multicast list - we cache all addresses so they won't
989 		 * change while HW is updated holding the command semaphor */
990 		netif_addr_lock_bh(dev);
991 		mlx4_en_cache_mclist(dev);
992 		netif_addr_unlock_bh(dev);
993 		list_for_each_entry(mclist, &priv->mc_list, list) {
994 			mcast_addr = mlx4_mac_to_u64(mclist->addr);
995 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
996 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
997 		}
998 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
999 					  0, MLX4_MCAST_ENABLE);
1000 		if (err)
1001 			en_err(priv, "Failed enabling multicast filter\n");
1002 
1003 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1004 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1005 			if (mclist->action == MCLIST_REM) {
1006 				/* detach this address and delete from list */
1007 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1008 				mc_list[5] = priv->port;
1009 				err = mlx4_multicast_detach(mdev->dev,
1010 							    &priv->rss_map.indir_qp,
1011 							    mc_list,
1012 							    MLX4_PROT_ETH,
1013 							    mclist->reg_id);
1014 				if (err)
1015 					en_err(priv, "Fail to detach multicast address\n");
1016 
1017 				if (mclist->tunnel_reg_id) {
1018 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1019 					if (err)
1020 						en_err(priv, "Failed to detach multicast address\n");
1021 				}
1022 
1023 				/* remove from list */
1024 				list_del(&mclist->list);
1025 				kfree(mclist);
1026 			} else if (mclist->action == MCLIST_ADD) {
1027 				/* attach the address */
1028 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1029 				/* needed for B0 steering support */
1030 				mc_list[5] = priv->port;
1031 				err = mlx4_multicast_attach(mdev->dev,
1032 							    &priv->rss_map.indir_qp,
1033 							    mc_list,
1034 							    priv->port, 0,
1035 							    MLX4_PROT_ETH,
1036 							    &mclist->reg_id);
1037 				if (err)
1038 					en_err(priv, "Fail to attach multicast address\n");
1039 
1040 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1041 							       &mclist->tunnel_reg_id);
1042 				if (err)
1043 					en_err(priv, "Failed to attach multicast address\n");
1044 			}
1045 		}
1046 	}
1047 }
1048 
1049 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1050 				 struct net_device *dev,
1051 				 struct mlx4_en_dev *mdev)
1052 {
1053 	struct netdev_hw_addr *ha;
1054 	struct mlx4_mac_entry *entry;
1055 	struct hlist_node *tmp;
1056 	bool found;
1057 	u64 mac;
1058 	int err = 0;
1059 	struct hlist_head *bucket;
1060 	unsigned int i;
1061 	int removed = 0;
1062 	u32 prev_flags;
1063 
1064 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1065 	 * since all modification code is protected by mdev->state_lock
1066 	 */
1067 
1068 	/* find what to remove */
1069 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1070 		bucket = &priv->mac_hash[i];
1071 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1072 			found = false;
1073 			netdev_for_each_uc_addr(ha, dev) {
1074 				if (ether_addr_equal_64bits(entry->mac,
1075 							    ha->addr)) {
1076 					found = true;
1077 					break;
1078 				}
1079 			}
1080 
1081 			/* MAC address of the port is not in uc list */
1082 			if (ether_addr_equal_64bits(entry->mac,
1083 						    priv->current_mac))
1084 				found = true;
1085 
1086 			if (!found) {
1087 				mac = mlx4_mac_to_u64(entry->mac);
1088 				mlx4_en_uc_steer_release(priv, entry->mac,
1089 							 priv->base_qpn,
1090 							 entry->reg_id);
1091 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1092 
1093 				hlist_del_rcu(&entry->hlist);
1094 				kfree_rcu(entry, rcu);
1095 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1096 				       entry->mac, priv->port);
1097 				++removed;
1098 			}
1099 		}
1100 	}
1101 
1102 	/* if we didn't remove anything, there is no use in trying to add
1103 	 * again once we are in a forced promisc mode state
1104 	 */
1105 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1106 		return;
1107 
1108 	prev_flags = priv->flags;
1109 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1110 
1111 	/* find what to add */
1112 	netdev_for_each_uc_addr(ha, dev) {
1113 		found = false;
1114 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1115 		hlist_for_each_entry(entry, bucket, hlist) {
1116 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1117 				found = true;
1118 				break;
1119 			}
1120 		}
1121 
1122 		if (!found) {
1123 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1124 			if (!entry) {
1125 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1126 				       ha->addr, priv->port);
1127 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1128 				break;
1129 			}
1130 			mac = mlx4_mac_to_u64(ha->addr);
1131 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1132 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1133 			if (err < 0) {
1134 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1135 				       ha->addr, priv->port, err);
1136 				kfree(entry);
1137 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1138 				break;
1139 			}
1140 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1141 						   &priv->base_qpn,
1142 						   &entry->reg_id);
1143 			if (err) {
1144 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1145 				       ha->addr, priv->port, err);
1146 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1147 				kfree(entry);
1148 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1149 				break;
1150 			} else {
1151 				unsigned int mac_hash;
1152 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1153 				       ha->addr, priv->port);
1154 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1155 				bucket = &priv->mac_hash[mac_hash];
1156 				hlist_add_head_rcu(&entry->hlist, bucket);
1157 			}
1158 		}
1159 	}
1160 
1161 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1162 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1163 			priv->port);
1164 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1165 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1166 			priv->port);
1167 	}
1168 }
1169 
1170 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1171 {
1172 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1173 						 rx_mode_task);
1174 	struct mlx4_en_dev *mdev = priv->mdev;
1175 	struct net_device *dev = priv->dev;
1176 
1177 	mutex_lock(&mdev->state_lock);
1178 	if (!mdev->device_up) {
1179 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1180 		goto out;
1181 	}
1182 	if (!priv->port_up) {
1183 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1184 		goto out;
1185 	}
1186 
1187 	if (!netif_carrier_ok(dev)) {
1188 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1189 			if (priv->port_state.link_state) {
1190 				priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1191 				netif_carrier_on(dev);
1192 				en_dbg(LINK, priv, "Link Up\n");
1193 			}
1194 		}
1195 	}
1196 
1197 	if (dev->priv_flags & IFF_UNICAST_FLT)
1198 		mlx4_en_do_uc_filter(priv, dev, mdev);
1199 
1200 	/* Promsicuous mode: disable all filters */
1201 	if ((dev->flags & IFF_PROMISC) ||
1202 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1203 		mlx4_en_set_promisc_mode(priv, mdev);
1204 		goto out;
1205 	}
1206 
1207 	/* Not in promiscuous mode */
1208 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1209 		mlx4_en_clear_promisc_mode(priv, mdev);
1210 
1211 	mlx4_en_do_multicast(priv, dev, mdev);
1212 out:
1213 	mutex_unlock(&mdev->state_lock);
1214 }
1215 
1216 #ifdef CONFIG_NET_POLL_CONTROLLER
1217 static void mlx4_en_netpoll(struct net_device *dev)
1218 {
1219 	struct mlx4_en_priv *priv = netdev_priv(dev);
1220 	struct mlx4_en_cq *cq;
1221 	int i;
1222 
1223 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1224 		cq = priv->tx_cq[TX][i];
1225 		napi_schedule(&cq->napi);
1226 	}
1227 }
1228 #endif
1229 
1230 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1231 {
1232 	u64 reg_id;
1233 	int err = 0;
1234 	int *qpn = &priv->base_qpn;
1235 	struct mlx4_mac_entry *entry;
1236 
1237 	err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1238 	if (err)
1239 		return err;
1240 
1241 	err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1242 				       &priv->tunnel_reg_id);
1243 	if (err)
1244 		goto tunnel_err;
1245 
1246 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1247 	if (!entry) {
1248 		err = -ENOMEM;
1249 		goto alloc_err;
1250 	}
1251 
1252 	memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1253 	memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1254 	entry->reg_id = reg_id;
1255 	hlist_add_head_rcu(&entry->hlist,
1256 			   &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1257 
1258 	return 0;
1259 
1260 alloc_err:
1261 	if (priv->tunnel_reg_id)
1262 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1263 
1264 tunnel_err:
1265 	mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1266 	return err;
1267 }
1268 
1269 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1270 {
1271 	u64 mac;
1272 	unsigned int i;
1273 	int qpn = priv->base_qpn;
1274 	struct hlist_head *bucket;
1275 	struct hlist_node *tmp;
1276 	struct mlx4_mac_entry *entry;
1277 
1278 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1279 		bucket = &priv->mac_hash[i];
1280 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1281 			mac = mlx4_mac_to_u64(entry->mac);
1282 			en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1283 			       entry->mac);
1284 			mlx4_en_uc_steer_release(priv, entry->mac,
1285 						 qpn, entry->reg_id);
1286 
1287 			mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1288 			hlist_del_rcu(&entry->hlist);
1289 			kfree_rcu(entry, rcu);
1290 		}
1291 	}
1292 
1293 	if (priv->tunnel_reg_id) {
1294 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1295 		priv->tunnel_reg_id = 0;
1296 	}
1297 }
1298 
1299 static void mlx4_en_tx_timeout(struct net_device *dev)
1300 {
1301 	struct mlx4_en_priv *priv = netdev_priv(dev);
1302 	struct mlx4_en_dev *mdev = priv->mdev;
1303 	int i;
1304 
1305 	if (netif_msg_timer(priv))
1306 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1307 
1308 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1309 		struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][i];
1310 
1311 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1312 			continue;
1313 		en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1314 			i, tx_ring->qpn, tx_ring->sp_cqn,
1315 			tx_ring->cons, tx_ring->prod);
1316 	}
1317 
1318 	priv->port_stats.tx_timeout++;
1319 	en_dbg(DRV, priv, "Scheduling watchdog\n");
1320 	queue_work(mdev->workqueue, &priv->watchdog_task);
1321 }
1322 
1323 
1324 static void
1325 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1326 {
1327 	struct mlx4_en_priv *priv = netdev_priv(dev);
1328 
1329 	spin_lock_bh(&priv->stats_lock);
1330 	mlx4_en_fold_software_stats(dev);
1331 	netdev_stats_to_stats64(stats, &dev->stats);
1332 	spin_unlock_bh(&priv->stats_lock);
1333 }
1334 
1335 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1336 {
1337 	struct mlx4_en_cq *cq;
1338 	int i, t;
1339 
1340 	/* If we haven't received a specific coalescing setting
1341 	 * (module param), we set the moderation parameters as follows:
1342 	 * - moder_cnt is set to the number of mtu sized packets to
1343 	 *   satisfy our coalescing target.
1344 	 * - moder_time is set to a fixed value.
1345 	 */
1346 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1347 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1348 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1349 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1350 	en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1351 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1352 
1353 	/* Setup cq moderation params */
1354 	for (i = 0; i < priv->rx_ring_num; i++) {
1355 		cq = priv->rx_cq[i];
1356 		cq->moder_cnt = priv->rx_frames;
1357 		cq->moder_time = priv->rx_usecs;
1358 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1359 		priv->last_moder_packets[i] = 0;
1360 		priv->last_moder_bytes[i] = 0;
1361 	}
1362 
1363 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1364 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1365 			cq = priv->tx_cq[t][i];
1366 			cq->moder_cnt = priv->tx_frames;
1367 			cq->moder_time = priv->tx_usecs;
1368 		}
1369 	}
1370 
1371 	/* Reset auto-moderation params */
1372 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1373 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1374 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1375 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1376 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1377 	priv->adaptive_rx_coal = 1;
1378 	priv->last_moder_jiffies = 0;
1379 	priv->last_moder_tx_packets = 0;
1380 }
1381 
1382 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1383 {
1384 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1385 	u32 pkt_rate_high, pkt_rate_low;
1386 	struct mlx4_en_cq *cq;
1387 	unsigned long packets;
1388 	unsigned long rate;
1389 	unsigned long avg_pkt_size;
1390 	unsigned long rx_packets;
1391 	unsigned long rx_bytes;
1392 	unsigned long rx_pkt_diff;
1393 	int moder_time;
1394 	int ring, err;
1395 
1396 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1397 		return;
1398 
1399 	pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1400 	pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1401 
1402 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1403 		rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1404 		rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1405 
1406 		rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1407 		packets = rx_pkt_diff;
1408 		rate = packets * HZ / period;
1409 		avg_pkt_size = packets ? (rx_bytes -
1410 				priv->last_moder_bytes[ring]) / packets : 0;
1411 
1412 		/* Apply auto-moderation only when packet rate
1413 		 * exceeds a rate that it matters */
1414 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1415 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1416 			if (rate <= pkt_rate_low)
1417 				moder_time = priv->rx_usecs_low;
1418 			else if (rate >= pkt_rate_high)
1419 				moder_time = priv->rx_usecs_high;
1420 			else
1421 				moder_time = (rate - pkt_rate_low) *
1422 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1423 					(pkt_rate_high - pkt_rate_low) +
1424 					priv->rx_usecs_low;
1425 		} else {
1426 			moder_time = priv->rx_usecs_low;
1427 		}
1428 
1429 		cq = priv->rx_cq[ring];
1430 		if (moder_time != priv->last_moder_time[ring] ||
1431 		    cq->moder_cnt != priv->rx_frames) {
1432 			priv->last_moder_time[ring] = moder_time;
1433 			cq->moder_time = moder_time;
1434 			cq->moder_cnt = priv->rx_frames;
1435 			err = mlx4_en_set_cq_moder(priv, cq);
1436 			if (err)
1437 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1438 				       ring);
1439 		}
1440 		priv->last_moder_packets[ring] = rx_packets;
1441 		priv->last_moder_bytes[ring] = rx_bytes;
1442 	}
1443 
1444 	priv->last_moder_jiffies = jiffies;
1445 }
1446 
1447 static void mlx4_en_do_get_stats(struct work_struct *work)
1448 {
1449 	struct delayed_work *delay = to_delayed_work(work);
1450 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1451 						 stats_task);
1452 	struct mlx4_en_dev *mdev = priv->mdev;
1453 	int err;
1454 
1455 	mutex_lock(&mdev->state_lock);
1456 	if (mdev->device_up) {
1457 		if (priv->port_up) {
1458 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1459 			if (err)
1460 				en_dbg(HW, priv, "Could not update stats\n");
1461 
1462 			mlx4_en_auto_moderation(priv);
1463 		}
1464 
1465 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1466 	}
1467 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1468 		mlx4_en_do_set_mac(priv, priv->current_mac);
1469 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1470 	}
1471 	mutex_unlock(&mdev->state_lock);
1472 }
1473 
1474 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1475  * periodically
1476  */
1477 static void mlx4_en_service_task(struct work_struct *work)
1478 {
1479 	struct delayed_work *delay = to_delayed_work(work);
1480 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1481 						 service_task);
1482 	struct mlx4_en_dev *mdev = priv->mdev;
1483 
1484 	mutex_lock(&mdev->state_lock);
1485 	if (mdev->device_up) {
1486 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1487 			mlx4_en_ptp_overflow_check(mdev);
1488 
1489 		mlx4_en_recover_from_oom(priv);
1490 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1491 				   SERVICE_TASK_DELAY);
1492 	}
1493 	mutex_unlock(&mdev->state_lock);
1494 }
1495 
1496 static void mlx4_en_linkstate(struct work_struct *work)
1497 {
1498 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1499 						 linkstate_task);
1500 	struct mlx4_en_dev *mdev = priv->mdev;
1501 	int linkstate = priv->link_state;
1502 
1503 	mutex_lock(&mdev->state_lock);
1504 	/* If observable port state changed set carrier state and
1505 	 * report to system log */
1506 	if (priv->last_link_state != linkstate) {
1507 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1508 			en_info(priv, "Link Down\n");
1509 			netif_carrier_off(priv->dev);
1510 		} else {
1511 			en_info(priv, "Link Up\n");
1512 			netif_carrier_on(priv->dev);
1513 		}
1514 	}
1515 	priv->last_link_state = linkstate;
1516 	mutex_unlock(&mdev->state_lock);
1517 }
1518 
1519 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1520 {
1521 	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1522 	int numa_node = priv->mdev->dev->numa_node;
1523 
1524 	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1525 		return -ENOMEM;
1526 
1527 	cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1528 			ring->affinity_mask);
1529 	return 0;
1530 }
1531 
1532 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1533 {
1534 	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1535 }
1536 
1537 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1538 				      int tx_ring_idx)
1539 {
1540 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1541 	int rr_index = tx_ring_idx;
1542 
1543 	tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1544 	tx_ring->recycle_ring = priv->rx_ring[rr_index];
1545 	en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1546 	       TX_XDP, tx_ring_idx, rr_index);
1547 }
1548 
1549 int mlx4_en_start_port(struct net_device *dev)
1550 {
1551 	struct mlx4_en_priv *priv = netdev_priv(dev);
1552 	struct mlx4_en_dev *mdev = priv->mdev;
1553 	struct mlx4_en_cq *cq;
1554 	struct mlx4_en_tx_ring *tx_ring;
1555 	int rx_index = 0;
1556 	int err = 0;
1557 	int i, t;
1558 	int j;
1559 	u8 mc_list[16] = {0};
1560 
1561 	if (priv->port_up) {
1562 		en_dbg(DRV, priv, "start port called while port already up\n");
1563 		return 0;
1564 	}
1565 
1566 	INIT_LIST_HEAD(&priv->mc_list);
1567 	INIT_LIST_HEAD(&priv->curr_list);
1568 	INIT_LIST_HEAD(&priv->ethtool_list);
1569 	memset(&priv->ethtool_rules[0], 0,
1570 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1571 
1572 	/* Calculate Rx buf size */
1573 	dev->mtu = min(dev->mtu, priv->max_mtu);
1574 	mlx4_en_calc_rx_buf(dev);
1575 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1576 
1577 	/* Configure rx cq's and rings */
1578 	err = mlx4_en_activate_rx_rings(priv);
1579 	if (err) {
1580 		en_err(priv, "Failed to activate RX rings\n");
1581 		return err;
1582 	}
1583 	for (i = 0; i < priv->rx_ring_num; i++) {
1584 		cq = priv->rx_cq[i];
1585 
1586 		err = mlx4_en_init_affinity_hint(priv, i);
1587 		if (err) {
1588 			en_err(priv, "Failed preparing IRQ affinity hint\n");
1589 			goto cq_err;
1590 		}
1591 
1592 		err = mlx4_en_activate_cq(priv, cq, i);
1593 		if (err) {
1594 			en_err(priv, "Failed activating Rx CQ\n");
1595 			mlx4_en_free_affinity_hint(priv, i);
1596 			goto cq_err;
1597 		}
1598 
1599 		for (j = 0; j < cq->size; j++) {
1600 			struct mlx4_cqe *cqe = NULL;
1601 
1602 			cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1603 			      priv->cqe_factor;
1604 			cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1605 		}
1606 
1607 		err = mlx4_en_set_cq_moder(priv, cq);
1608 		if (err) {
1609 			en_err(priv, "Failed setting cq moderation parameters\n");
1610 			mlx4_en_deactivate_cq(priv, cq);
1611 			mlx4_en_free_affinity_hint(priv, i);
1612 			goto cq_err;
1613 		}
1614 		mlx4_en_arm_cq(priv, cq);
1615 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1616 		++rx_index;
1617 	}
1618 
1619 	/* Set qp number */
1620 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1621 	err = mlx4_en_get_qp(priv);
1622 	if (err) {
1623 		en_err(priv, "Failed getting eth qp\n");
1624 		goto cq_err;
1625 	}
1626 	mdev->mac_removed[priv->port] = 0;
1627 
1628 	priv->counter_index =
1629 			mlx4_get_default_counter_index(mdev->dev, priv->port);
1630 
1631 	err = mlx4_en_config_rss_steer(priv);
1632 	if (err) {
1633 		en_err(priv, "Failed configuring rss steering\n");
1634 		goto mac_err;
1635 	}
1636 
1637 	err = mlx4_en_create_drop_qp(priv);
1638 	if (err)
1639 		goto rss_err;
1640 
1641 	/* Configure tx cq's and rings */
1642 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1643 		u8 num_tx_rings_p_up = t == TX ?
1644 			priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1645 
1646 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1647 			/* Configure cq */
1648 			cq = priv->tx_cq[t][i];
1649 			err = mlx4_en_activate_cq(priv, cq, i);
1650 			if (err) {
1651 				en_err(priv, "Failed allocating Tx CQ\n");
1652 				goto tx_err;
1653 			}
1654 			err = mlx4_en_set_cq_moder(priv, cq);
1655 			if (err) {
1656 				en_err(priv, "Failed setting cq moderation parameters\n");
1657 				mlx4_en_deactivate_cq(priv, cq);
1658 				goto tx_err;
1659 			}
1660 			en_dbg(DRV, priv,
1661 			       "Resetting index of collapsed CQ:%d to -1\n", i);
1662 			cq->buf->wqe_index = cpu_to_be16(0xffff);
1663 
1664 			/* Configure ring */
1665 			tx_ring = priv->tx_ring[t][i];
1666 			err = mlx4_en_activate_tx_ring(priv, tx_ring,
1667 						       cq->mcq.cqn,
1668 						       i / num_tx_rings_p_up);
1669 			if (err) {
1670 				en_err(priv, "Failed allocating Tx ring\n");
1671 				mlx4_en_deactivate_cq(priv, cq);
1672 				goto tx_err;
1673 			}
1674 			if (t != TX_XDP) {
1675 				tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1676 				tx_ring->recycle_ring = NULL;
1677 			} else {
1678 				mlx4_en_init_recycle_ring(priv, i);
1679 			}
1680 
1681 			/* Arm CQ for TX completions */
1682 			mlx4_en_arm_cq(priv, cq);
1683 
1684 			/* Set initial ownership of all Tx TXBBs to SW (1) */
1685 			for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1686 				*((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1687 		}
1688 	}
1689 
1690 	/* Configure port */
1691 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1692 				    priv->rx_skb_size + ETH_FCS_LEN,
1693 				    priv->prof->tx_pause,
1694 				    priv->prof->tx_ppp,
1695 				    priv->prof->rx_pause,
1696 				    priv->prof->rx_ppp);
1697 	if (err) {
1698 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1699 		       priv->port, err);
1700 		goto tx_err;
1701 	}
1702 
1703 	err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1704 	if (err) {
1705 		en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1706 		       dev->mtu, priv->port, err);
1707 		goto tx_err;
1708 	}
1709 
1710 	/* Set default qp number */
1711 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1712 	if (err) {
1713 		en_err(priv, "Failed setting default qp numbers\n");
1714 		goto tx_err;
1715 	}
1716 
1717 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1718 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1719 		if (err) {
1720 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1721 			       err);
1722 			goto tx_err;
1723 		}
1724 	}
1725 
1726 	/* Init port */
1727 	en_dbg(HW, priv, "Initializing port\n");
1728 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1729 	if (err) {
1730 		en_err(priv, "Failed Initializing port\n");
1731 		goto tx_err;
1732 	}
1733 
1734 	/* Set Unicast and VXLAN steering rules */
1735 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1736 	    mlx4_en_set_rss_steer_rules(priv))
1737 		mlx4_warn(mdev, "Failed setting steering rules\n");
1738 
1739 	/* Attach rx QP to bradcast address */
1740 	eth_broadcast_addr(&mc_list[10]);
1741 	mc_list[5] = priv->port; /* needed for B0 steering support */
1742 	if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1743 				  priv->port, 0, MLX4_PROT_ETH,
1744 				  &priv->broadcast_id))
1745 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1746 
1747 	/* Must redo promiscuous mode setup. */
1748 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1749 
1750 	/* Schedule multicast task to populate multicast list */
1751 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1752 
1753 	if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1754 		udp_tunnel_get_rx_info(dev);
1755 
1756 	priv->port_up = true;
1757 
1758 	/* Process all completions if exist to prevent
1759 	 * the queues freezing if they are full
1760 	 */
1761 	for (i = 0; i < priv->rx_ring_num; i++) {
1762 		local_bh_disable();
1763 		napi_schedule(&priv->rx_cq[i]->napi);
1764 		local_bh_enable();
1765 	}
1766 
1767 	netif_tx_start_all_queues(dev);
1768 	netif_device_attach(dev);
1769 
1770 	return 0;
1771 
1772 tx_err:
1773 	if (t == MLX4_EN_NUM_TX_TYPES) {
1774 		t--;
1775 		i = priv->tx_ring_num[t];
1776 	}
1777 	while (t >= 0) {
1778 		while (i--) {
1779 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1780 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1781 		}
1782 		if (!t--)
1783 			break;
1784 		i = priv->tx_ring_num[t];
1785 	}
1786 	mlx4_en_destroy_drop_qp(priv);
1787 rss_err:
1788 	mlx4_en_release_rss_steer(priv);
1789 mac_err:
1790 	mlx4_en_put_qp(priv);
1791 cq_err:
1792 	while (rx_index--) {
1793 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1794 		mlx4_en_free_affinity_hint(priv, rx_index);
1795 	}
1796 	for (i = 0; i < priv->rx_ring_num; i++)
1797 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1798 
1799 	return err; /* need to close devices */
1800 }
1801 
1802 
1803 void mlx4_en_stop_port(struct net_device *dev, int detach)
1804 {
1805 	struct mlx4_en_priv *priv = netdev_priv(dev);
1806 	struct mlx4_en_dev *mdev = priv->mdev;
1807 	struct mlx4_en_mc_list *mclist, *tmp;
1808 	struct ethtool_flow_id *flow, *tmp_flow;
1809 	int i, t;
1810 	u8 mc_list[16] = {0};
1811 
1812 	if (!priv->port_up) {
1813 		en_dbg(DRV, priv, "stop port called while port already down\n");
1814 		return;
1815 	}
1816 
1817 	/* close port*/
1818 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1819 
1820 	/* Synchronize with tx routine */
1821 	netif_tx_lock_bh(dev);
1822 	if (detach)
1823 		netif_device_detach(dev);
1824 	netif_tx_stop_all_queues(dev);
1825 	netif_tx_unlock_bh(dev);
1826 
1827 	netif_tx_disable(dev);
1828 
1829 	spin_lock_bh(&priv->stats_lock);
1830 	mlx4_en_fold_software_stats(dev);
1831 	/* Set port as not active */
1832 	priv->port_up = false;
1833 	spin_unlock_bh(&priv->stats_lock);
1834 
1835 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1836 
1837 	/* Promsicuous mode */
1838 	if (mdev->dev->caps.steering_mode ==
1839 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1840 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1841 				 MLX4_EN_FLAG_MC_PROMISC);
1842 		mlx4_flow_steer_promisc_remove(mdev->dev,
1843 					       priv->port,
1844 					       MLX4_FS_ALL_DEFAULT);
1845 		mlx4_flow_steer_promisc_remove(mdev->dev,
1846 					       priv->port,
1847 					       MLX4_FS_MC_DEFAULT);
1848 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1849 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1850 
1851 		/* Disable promiscouos mode */
1852 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1853 					    priv->port);
1854 
1855 		/* Disable Multicast promisc */
1856 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1857 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1858 						      priv->port);
1859 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1860 		}
1861 	}
1862 
1863 	/* Detach All multicasts */
1864 	eth_broadcast_addr(&mc_list[10]);
1865 	mc_list[5] = priv->port; /* needed for B0 steering support */
1866 	mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1867 			      MLX4_PROT_ETH, priv->broadcast_id);
1868 	list_for_each_entry(mclist, &priv->curr_list, list) {
1869 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1870 		mc_list[5] = priv->port;
1871 		mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1872 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1873 		if (mclist->tunnel_reg_id)
1874 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1875 	}
1876 	mlx4_en_clear_list(dev);
1877 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1878 		list_del(&mclist->list);
1879 		kfree(mclist);
1880 	}
1881 
1882 	/* Flush multicast filter */
1883 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1884 
1885 	/* Remove flow steering rules for the port*/
1886 	if (mdev->dev->caps.steering_mode ==
1887 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1888 		ASSERT_RTNL();
1889 		list_for_each_entry_safe(flow, tmp_flow,
1890 					 &priv->ethtool_list, list) {
1891 			mlx4_flow_detach(mdev->dev, flow->id);
1892 			list_del(&flow->list);
1893 		}
1894 	}
1895 
1896 	mlx4_en_destroy_drop_qp(priv);
1897 
1898 	/* Free TX Rings */
1899 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1900 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1901 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1902 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1903 		}
1904 	}
1905 	msleep(10);
1906 
1907 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1908 		for (i = 0; i < priv->tx_ring_num[t]; i++)
1909 			mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1910 
1911 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1912 		mlx4_en_delete_rss_steer_rules(priv);
1913 
1914 	/* Free RSS qps */
1915 	mlx4_en_release_rss_steer(priv);
1916 
1917 	/* Unregister Mac address for the port */
1918 	mlx4_en_put_qp(priv);
1919 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1920 		mdev->mac_removed[priv->port] = 1;
1921 
1922 	/* Free RX Rings */
1923 	for (i = 0; i < priv->rx_ring_num; i++) {
1924 		struct mlx4_en_cq *cq = priv->rx_cq[i];
1925 
1926 		napi_synchronize(&cq->napi);
1927 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1928 		mlx4_en_deactivate_cq(priv, cq);
1929 
1930 		mlx4_en_free_affinity_hint(priv, i);
1931 	}
1932 }
1933 
1934 static void mlx4_en_restart(struct work_struct *work)
1935 {
1936 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1937 						 watchdog_task);
1938 	struct mlx4_en_dev *mdev = priv->mdev;
1939 	struct net_device *dev = priv->dev;
1940 
1941 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1942 
1943 	rtnl_lock();
1944 	mutex_lock(&mdev->state_lock);
1945 	if (priv->port_up) {
1946 		mlx4_en_stop_port(dev, 1);
1947 		if (mlx4_en_start_port(dev))
1948 			en_err(priv, "Failed restarting port %d\n", priv->port);
1949 	}
1950 	mutex_unlock(&mdev->state_lock);
1951 	rtnl_unlock();
1952 }
1953 
1954 static void mlx4_en_clear_stats(struct net_device *dev)
1955 {
1956 	struct mlx4_en_priv *priv = netdev_priv(dev);
1957 	struct mlx4_en_dev *mdev = priv->mdev;
1958 	struct mlx4_en_tx_ring **tx_ring;
1959 	int i;
1960 
1961 	if (!mlx4_is_slave(mdev->dev))
1962 		if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1963 			en_dbg(HW, priv, "Failed dumping statistics\n");
1964 
1965 	memset(&priv->pstats, 0, sizeof(priv->pstats));
1966 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1967 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1968 	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
1969 	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
1970 	memset(&priv->rx_priority_flowstats, 0,
1971 	       sizeof(priv->rx_priority_flowstats));
1972 	memset(&priv->tx_priority_flowstats, 0,
1973 	       sizeof(priv->tx_priority_flowstats));
1974 	memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
1975 
1976 	tx_ring = priv->tx_ring[TX];
1977 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1978 		tx_ring[i]->bytes = 0;
1979 		tx_ring[i]->packets = 0;
1980 		tx_ring[i]->tx_csum = 0;
1981 		tx_ring[i]->tx_dropped = 0;
1982 		tx_ring[i]->queue_stopped = 0;
1983 		tx_ring[i]->wake_queue = 0;
1984 		tx_ring[i]->tso_packets = 0;
1985 		tx_ring[i]->xmit_more = 0;
1986 	}
1987 	for (i = 0; i < priv->rx_ring_num; i++) {
1988 		priv->rx_ring[i]->bytes = 0;
1989 		priv->rx_ring[i]->packets = 0;
1990 		priv->rx_ring[i]->csum_ok = 0;
1991 		priv->rx_ring[i]->csum_none = 0;
1992 		priv->rx_ring[i]->csum_complete = 0;
1993 	}
1994 }
1995 
1996 static int mlx4_en_open(struct net_device *dev)
1997 {
1998 	struct mlx4_en_priv *priv = netdev_priv(dev);
1999 	struct mlx4_en_dev *mdev = priv->mdev;
2000 	int err = 0;
2001 
2002 	mutex_lock(&mdev->state_lock);
2003 
2004 	if (!mdev->device_up) {
2005 		en_err(priv, "Cannot open - device down/disabled\n");
2006 		err = -EBUSY;
2007 		goto out;
2008 	}
2009 
2010 	/* Reset HW statistics and SW counters */
2011 	mlx4_en_clear_stats(dev);
2012 
2013 	err = mlx4_en_start_port(dev);
2014 	if (err)
2015 		en_err(priv, "Failed starting port:%d\n", priv->port);
2016 
2017 out:
2018 	mutex_unlock(&mdev->state_lock);
2019 	return err;
2020 }
2021 
2022 
2023 static int mlx4_en_close(struct net_device *dev)
2024 {
2025 	struct mlx4_en_priv *priv = netdev_priv(dev);
2026 	struct mlx4_en_dev *mdev = priv->mdev;
2027 
2028 	en_dbg(IFDOWN, priv, "Close port called\n");
2029 
2030 	mutex_lock(&mdev->state_lock);
2031 
2032 	mlx4_en_stop_port(dev, 0);
2033 	netif_carrier_off(dev);
2034 
2035 	mutex_unlock(&mdev->state_lock);
2036 	return 0;
2037 }
2038 
2039 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2040 {
2041 	int i, t;
2042 
2043 #ifdef CONFIG_RFS_ACCEL
2044 	priv->dev->rx_cpu_rmap = NULL;
2045 #endif
2046 
2047 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2048 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2049 			if (priv->tx_ring[t] && priv->tx_ring[t][i])
2050 				mlx4_en_destroy_tx_ring(priv,
2051 							&priv->tx_ring[t][i]);
2052 			if (priv->tx_cq[t] && priv->tx_cq[t][i])
2053 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2054 		}
2055 		kfree(priv->tx_ring[t]);
2056 		kfree(priv->tx_cq[t]);
2057 	}
2058 
2059 	for (i = 0; i < priv->rx_ring_num; i++) {
2060 		if (priv->rx_ring[i])
2061 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2062 				priv->prof->rx_ring_size, priv->stride);
2063 		if (priv->rx_cq[i])
2064 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2065 	}
2066 
2067 }
2068 
2069 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2070 {
2071 	struct mlx4_en_port_profile *prof = priv->prof;
2072 	int i, t;
2073 	int node;
2074 
2075 	/* Create tx Rings */
2076 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2077 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2078 			node = cpu_to_node(i % num_online_cpus());
2079 			if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2080 					      prof->tx_ring_size, i, t, node))
2081 				goto err;
2082 
2083 			if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2084 						   prof->tx_ring_size,
2085 						   TXBB_SIZE, node, i))
2086 				goto err;
2087 		}
2088 	}
2089 
2090 	/* Create rx Rings */
2091 	for (i = 0; i < priv->rx_ring_num; i++) {
2092 		node = cpu_to_node(i % num_online_cpus());
2093 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2094 				      prof->rx_ring_size, i, RX, node))
2095 			goto err;
2096 
2097 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2098 					   prof->rx_ring_size, priv->stride,
2099 					   node))
2100 			goto err;
2101 	}
2102 
2103 #ifdef CONFIG_RFS_ACCEL
2104 	priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2105 #endif
2106 
2107 	return 0;
2108 
2109 err:
2110 	en_err(priv, "Failed to allocate NIC resources\n");
2111 	for (i = 0; i < priv->rx_ring_num; i++) {
2112 		if (priv->rx_ring[i])
2113 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2114 						prof->rx_ring_size,
2115 						priv->stride);
2116 		if (priv->rx_cq[i])
2117 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2118 	}
2119 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2120 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2121 			if (priv->tx_ring[t][i])
2122 				mlx4_en_destroy_tx_ring(priv,
2123 							&priv->tx_ring[t][i]);
2124 			if (priv->tx_cq[t][i])
2125 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2126 		}
2127 	}
2128 	return -ENOMEM;
2129 }
2130 
2131 
2132 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2133 			     struct mlx4_en_priv *src,
2134 			     struct mlx4_en_port_profile *prof)
2135 {
2136 	int t;
2137 
2138 	memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2139 	       sizeof(dst->hwtstamp_config));
2140 	dst->num_tx_rings_p_up = src->mdev->profile.num_tx_rings_p_up;
2141 	dst->rx_ring_num = prof->rx_ring_num;
2142 	dst->flags = prof->flags;
2143 	dst->mdev = src->mdev;
2144 	dst->port = src->port;
2145 	dst->dev = src->dev;
2146 	dst->prof = prof;
2147 	dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2148 					 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2149 
2150 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2151 		dst->tx_ring_num[t] = prof->tx_ring_num[t];
2152 		if (!dst->tx_ring_num[t])
2153 			continue;
2154 
2155 		dst->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
2156 					  MAX_TX_RINGS, GFP_KERNEL);
2157 		if (!dst->tx_ring[t])
2158 			goto err_free_tx;
2159 
2160 		dst->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
2161 					MAX_TX_RINGS, GFP_KERNEL);
2162 		if (!dst->tx_cq[t]) {
2163 			kfree(dst->tx_ring[t]);
2164 			goto err_free_tx;
2165 		}
2166 	}
2167 
2168 	return 0;
2169 
2170 err_free_tx:
2171 	while (t--) {
2172 		kfree(dst->tx_ring[t]);
2173 		kfree(dst->tx_cq[t]);
2174 	}
2175 	return -ENOMEM;
2176 }
2177 
2178 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2179 				struct mlx4_en_priv *src)
2180 {
2181 	int t;
2182 	memcpy(dst->rx_ring, src->rx_ring,
2183 	       sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2184 	memcpy(dst->rx_cq, src->rx_cq,
2185 	       sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2186 	memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2187 	       sizeof(dst->hwtstamp_config));
2188 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2189 		dst->tx_ring_num[t] = src->tx_ring_num[t];
2190 		dst->tx_ring[t] = src->tx_ring[t];
2191 		dst->tx_cq[t] = src->tx_cq[t];
2192 	}
2193 	dst->rx_ring_num = src->rx_ring_num;
2194 	memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2195 }
2196 
2197 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2198 				struct mlx4_en_priv *tmp,
2199 				struct mlx4_en_port_profile *prof,
2200 				bool carry_xdp_prog)
2201 {
2202 	struct bpf_prog *xdp_prog;
2203 	int i, t;
2204 
2205 	mlx4_en_copy_priv(tmp, priv, prof);
2206 
2207 	if (mlx4_en_alloc_resources(tmp)) {
2208 		en_warn(priv,
2209 			"%s: Resource allocation failed, using previous configuration\n",
2210 			__func__);
2211 		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2212 			kfree(tmp->tx_ring[t]);
2213 			kfree(tmp->tx_cq[t]);
2214 		}
2215 		return -ENOMEM;
2216 	}
2217 
2218 	/* All rx_rings has the same xdp_prog.  Pick the first one. */
2219 	xdp_prog = rcu_dereference_protected(
2220 		priv->rx_ring[0]->xdp_prog,
2221 		lockdep_is_held(&priv->mdev->state_lock));
2222 
2223 	if (xdp_prog && carry_xdp_prog) {
2224 		xdp_prog = bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2225 		if (IS_ERR(xdp_prog)) {
2226 			mlx4_en_free_resources(tmp);
2227 			return PTR_ERR(xdp_prog);
2228 		}
2229 		for (i = 0; i < tmp->rx_ring_num; i++)
2230 			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2231 					   xdp_prog);
2232 	}
2233 
2234 	return 0;
2235 }
2236 
2237 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2238 				    struct mlx4_en_priv *tmp)
2239 {
2240 	mlx4_en_free_resources(priv);
2241 	mlx4_en_update_priv(priv, tmp);
2242 }
2243 
2244 void mlx4_en_destroy_netdev(struct net_device *dev)
2245 {
2246 	struct mlx4_en_priv *priv = netdev_priv(dev);
2247 	struct mlx4_en_dev *mdev = priv->mdev;
2248 
2249 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2250 
2251 	/* Unregister device - this will close the port if it was up */
2252 	if (priv->registered) {
2253 		devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2254 							      priv->port));
2255 		unregister_netdev(dev);
2256 	}
2257 
2258 	if (priv->allocated)
2259 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2260 
2261 	cancel_delayed_work(&priv->stats_task);
2262 	cancel_delayed_work(&priv->service_task);
2263 	/* flush any pending task for this netdev */
2264 	flush_workqueue(mdev->workqueue);
2265 
2266 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2267 		mlx4_en_remove_timestamp(mdev);
2268 
2269 	/* Detach the netdev so tasks would not attempt to access it */
2270 	mutex_lock(&mdev->state_lock);
2271 	mdev->pndev[priv->port] = NULL;
2272 	mdev->upper[priv->port] = NULL;
2273 
2274 #ifdef CONFIG_RFS_ACCEL
2275 	mlx4_en_cleanup_filters(priv);
2276 #endif
2277 
2278 	mlx4_en_free_resources(priv);
2279 	mutex_unlock(&mdev->state_lock);
2280 
2281 	free_netdev(dev);
2282 }
2283 
2284 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2285 {
2286 	struct mlx4_en_priv *priv = netdev_priv(dev);
2287 
2288 	if (mtu > MLX4_EN_MAX_XDP_MTU) {
2289 		en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2290 		       mtu, MLX4_EN_MAX_XDP_MTU);
2291 		return false;
2292 	}
2293 
2294 	return true;
2295 }
2296 
2297 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2298 {
2299 	struct mlx4_en_priv *priv = netdev_priv(dev);
2300 	struct mlx4_en_dev *mdev = priv->mdev;
2301 	int err = 0;
2302 
2303 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2304 		 dev->mtu, new_mtu);
2305 
2306 	if (priv->tx_ring_num[TX_XDP] &&
2307 	    !mlx4_en_check_xdp_mtu(dev, new_mtu))
2308 		return -EOPNOTSUPP;
2309 
2310 	dev->mtu = new_mtu;
2311 
2312 	if (netif_running(dev)) {
2313 		mutex_lock(&mdev->state_lock);
2314 		if (!mdev->device_up) {
2315 			/* NIC is probably restarting - let watchdog task reset
2316 			 * the port */
2317 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2318 		} else {
2319 			mlx4_en_stop_port(dev, 1);
2320 			err = mlx4_en_start_port(dev);
2321 			if (err) {
2322 				en_err(priv, "Failed restarting port:%d\n",
2323 					 priv->port);
2324 				queue_work(mdev->workqueue, &priv->watchdog_task);
2325 			}
2326 		}
2327 		mutex_unlock(&mdev->state_lock);
2328 	}
2329 	return 0;
2330 }
2331 
2332 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2333 {
2334 	struct mlx4_en_priv *priv = netdev_priv(dev);
2335 	struct mlx4_en_dev *mdev = priv->mdev;
2336 	struct hwtstamp_config config;
2337 
2338 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2339 		return -EFAULT;
2340 
2341 	/* reserved for future extensions */
2342 	if (config.flags)
2343 		return -EINVAL;
2344 
2345 	/* device doesn't support time stamping */
2346 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2347 		return -EINVAL;
2348 
2349 	/* TX HW timestamp */
2350 	switch (config.tx_type) {
2351 	case HWTSTAMP_TX_OFF:
2352 	case HWTSTAMP_TX_ON:
2353 		break;
2354 	default:
2355 		return -ERANGE;
2356 	}
2357 
2358 	/* RX HW timestamp */
2359 	switch (config.rx_filter) {
2360 	case HWTSTAMP_FILTER_NONE:
2361 		break;
2362 	case HWTSTAMP_FILTER_ALL:
2363 	case HWTSTAMP_FILTER_SOME:
2364 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2365 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2366 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2367 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2368 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2369 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2370 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2371 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2372 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2373 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2374 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2375 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2376 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2377 		break;
2378 	default:
2379 		return -ERANGE;
2380 	}
2381 
2382 	if (mlx4_en_reset_config(dev, config, dev->features)) {
2383 		config.tx_type = HWTSTAMP_TX_OFF;
2384 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2385 	}
2386 
2387 	return copy_to_user(ifr->ifr_data, &config,
2388 			    sizeof(config)) ? -EFAULT : 0;
2389 }
2390 
2391 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2392 {
2393 	struct mlx4_en_priv *priv = netdev_priv(dev);
2394 
2395 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2396 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2397 }
2398 
2399 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2400 {
2401 	switch (cmd) {
2402 	case SIOCSHWTSTAMP:
2403 		return mlx4_en_hwtstamp_set(dev, ifr);
2404 	case SIOCGHWTSTAMP:
2405 		return mlx4_en_hwtstamp_get(dev, ifr);
2406 	default:
2407 		return -EOPNOTSUPP;
2408 	}
2409 }
2410 
2411 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2412 					      netdev_features_t features)
2413 {
2414 	struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2415 	struct mlx4_en_dev *mdev = en_priv->mdev;
2416 
2417 	/* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2418 	 * enable/disable make sure S-TAG flag is always in same state as
2419 	 * C-TAG.
2420 	 */
2421 	if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2422 	    !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2423 		features |= NETIF_F_HW_VLAN_STAG_RX;
2424 	else
2425 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2426 
2427 	return features;
2428 }
2429 
2430 static int mlx4_en_set_features(struct net_device *netdev,
2431 		netdev_features_t features)
2432 {
2433 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2434 	bool reset = false;
2435 	int ret = 0;
2436 
2437 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2438 		en_info(priv, "Turn %s RX-FCS\n",
2439 			(features & NETIF_F_RXFCS) ? "ON" : "OFF");
2440 		reset = true;
2441 	}
2442 
2443 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2444 		u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2445 
2446 		en_info(priv, "Turn %s RX-ALL\n",
2447 			ignore_fcs_value ? "ON" : "OFF");
2448 		ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2449 					      priv->port, ignore_fcs_value);
2450 		if (ret)
2451 			return ret;
2452 	}
2453 
2454 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2455 		en_info(priv, "Turn %s RX vlan strip offload\n",
2456 			(features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2457 		reset = true;
2458 	}
2459 
2460 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2461 		en_info(priv, "Turn %s TX vlan strip offload\n",
2462 			(features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2463 
2464 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2465 		en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2466 			(features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2467 
2468 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2469 		en_info(priv, "Turn %s loopback\n",
2470 			(features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2471 		mlx4_en_update_loopback_state(netdev, features);
2472 	}
2473 
2474 	if (reset) {
2475 		ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2476 					   features);
2477 		if (ret)
2478 			return ret;
2479 	}
2480 
2481 	return 0;
2482 }
2483 
2484 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2485 {
2486 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2487 	struct mlx4_en_dev *mdev = en_priv->mdev;
2488 	u64 mac_u64 = mlx4_mac_to_u64(mac);
2489 
2490 	if (is_multicast_ether_addr(mac))
2491 		return -EINVAL;
2492 
2493 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2494 }
2495 
2496 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2497 			       __be16 vlan_proto)
2498 {
2499 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2500 	struct mlx4_en_dev *mdev = en_priv->mdev;
2501 
2502 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2503 				vlan_proto);
2504 }
2505 
2506 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2507 			       int max_tx_rate)
2508 {
2509 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2510 	struct mlx4_en_dev *mdev = en_priv->mdev;
2511 
2512 	return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2513 				max_tx_rate);
2514 }
2515 
2516 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2517 {
2518 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2519 	struct mlx4_en_dev *mdev = en_priv->mdev;
2520 
2521 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2522 }
2523 
2524 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2525 {
2526 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2527 	struct mlx4_en_dev *mdev = en_priv->mdev;
2528 
2529 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2530 }
2531 
2532 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2533 {
2534 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2535 	struct mlx4_en_dev *mdev = en_priv->mdev;
2536 
2537 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2538 }
2539 
2540 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2541 				struct ifla_vf_stats *vf_stats)
2542 {
2543 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2544 	struct mlx4_en_dev *mdev = en_priv->mdev;
2545 
2546 	return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2547 }
2548 
2549 #define PORT_ID_BYTE_LEN 8
2550 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2551 				    struct netdev_phys_item_id *ppid)
2552 {
2553 	struct mlx4_en_priv *priv = netdev_priv(dev);
2554 	struct mlx4_dev *mdev = priv->mdev->dev;
2555 	int i;
2556 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2557 
2558 	if (!phys_port_id)
2559 		return -EOPNOTSUPP;
2560 
2561 	ppid->id_len = sizeof(phys_port_id);
2562 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2563 		ppid->id[i] =  phys_port_id & 0xff;
2564 		phys_port_id >>= 8;
2565 	}
2566 	return 0;
2567 }
2568 
2569 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2570 {
2571 	int ret;
2572 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2573 						 vxlan_add_task);
2574 
2575 	ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2576 	if (ret)
2577 		goto out;
2578 
2579 	ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2580 				  VXLAN_STEER_BY_OUTER_MAC, 1);
2581 out:
2582 	if (ret) {
2583 		en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2584 		return;
2585 	}
2586 
2587 	/* set offloads */
2588 	priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2589 				      NETIF_F_RXCSUM |
2590 				      NETIF_F_TSO | NETIF_F_TSO6 |
2591 				      NETIF_F_GSO_UDP_TUNNEL |
2592 				      NETIF_F_GSO_UDP_TUNNEL_CSUM |
2593 				      NETIF_F_GSO_PARTIAL;
2594 }
2595 
2596 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2597 {
2598 	int ret;
2599 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2600 						 vxlan_del_task);
2601 	/* unset offloads */
2602 	priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2603 					NETIF_F_RXCSUM |
2604 					NETIF_F_TSO | NETIF_F_TSO6 |
2605 					NETIF_F_GSO_UDP_TUNNEL |
2606 					NETIF_F_GSO_UDP_TUNNEL_CSUM |
2607 					NETIF_F_GSO_PARTIAL);
2608 
2609 	ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2610 				  VXLAN_STEER_BY_OUTER_MAC, 0);
2611 	if (ret)
2612 		en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2613 
2614 	priv->vxlan_port = 0;
2615 }
2616 
2617 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2618 				   struct udp_tunnel_info *ti)
2619 {
2620 	struct mlx4_en_priv *priv = netdev_priv(dev);
2621 	__be16 port = ti->port;
2622 	__be16 current_port;
2623 
2624 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2625 		return;
2626 
2627 	if (ti->sa_family != AF_INET)
2628 		return;
2629 
2630 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2631 		return;
2632 
2633 	current_port = priv->vxlan_port;
2634 	if (current_port && current_port != port) {
2635 		en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2636 			ntohs(current_port), ntohs(port));
2637 		return;
2638 	}
2639 
2640 	priv->vxlan_port = port;
2641 	queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2642 }
2643 
2644 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2645 				   struct udp_tunnel_info *ti)
2646 {
2647 	struct mlx4_en_priv *priv = netdev_priv(dev);
2648 	__be16 port = ti->port;
2649 	__be16 current_port;
2650 
2651 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2652 		return;
2653 
2654 	if (ti->sa_family != AF_INET)
2655 		return;
2656 
2657 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2658 		return;
2659 
2660 	current_port = priv->vxlan_port;
2661 	if (current_port != port) {
2662 		en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2663 		return;
2664 	}
2665 
2666 	queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2667 }
2668 
2669 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2670 						struct net_device *dev,
2671 						netdev_features_t features)
2672 {
2673 	features = vlan_features_check(skb, features);
2674 	features = vxlan_features_check(skb, features);
2675 
2676 	/* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2677 	 * support inner IPv6 checksums and segmentation so  we need to
2678 	 * strip that feature if this is an IPv6 encapsulated frame.
2679 	 */
2680 	if (skb->encapsulation &&
2681 	    (skb->ip_summed == CHECKSUM_PARTIAL)) {
2682 		struct mlx4_en_priv *priv = netdev_priv(dev);
2683 
2684 		if (!priv->vxlan_port ||
2685 		    (ip_hdr(skb)->version != 4) ||
2686 		    (udp_hdr(skb)->dest != priv->vxlan_port))
2687 			features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2688 	}
2689 
2690 	return features;
2691 }
2692 
2693 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2694 {
2695 	struct mlx4_en_priv *priv = netdev_priv(dev);
2696 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2697 	struct mlx4_update_qp_params params;
2698 	int err;
2699 
2700 	if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2701 		return -EOPNOTSUPP;
2702 
2703 	/* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2704 	if (maxrate >> 12) {
2705 		params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2706 		params.rate_val  = maxrate / 1000;
2707 	} else if (maxrate) {
2708 		params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2709 		params.rate_val  = maxrate;
2710 	} else { /* zero serves to revoke the QP rate-limitation */
2711 		params.rate_unit = 0;
2712 		params.rate_val  = 0;
2713 	}
2714 
2715 	err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2716 			     &params);
2717 	return err;
2718 }
2719 
2720 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2721 {
2722 	struct mlx4_en_priv *priv = netdev_priv(dev);
2723 	struct mlx4_en_dev *mdev = priv->mdev;
2724 	struct mlx4_en_port_profile new_prof;
2725 	struct bpf_prog *old_prog;
2726 	struct mlx4_en_priv *tmp;
2727 	int tx_changed = 0;
2728 	int xdp_ring_num;
2729 	int port_up = 0;
2730 	int err;
2731 	int i;
2732 
2733 	xdp_ring_num = prog ? priv->rx_ring_num : 0;
2734 
2735 	/* No need to reconfigure buffers when simply swapping the
2736 	 * program for a new one.
2737 	 */
2738 	if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2739 		if (prog) {
2740 			prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2741 			if (IS_ERR(prog))
2742 				return PTR_ERR(prog);
2743 		}
2744 		mutex_lock(&mdev->state_lock);
2745 		for (i = 0; i < priv->rx_ring_num; i++) {
2746 			old_prog = rcu_dereference_protected(
2747 					priv->rx_ring[i]->xdp_prog,
2748 					lockdep_is_held(&mdev->state_lock));
2749 			rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2750 			if (old_prog)
2751 				bpf_prog_put(old_prog);
2752 		}
2753 		mutex_unlock(&mdev->state_lock);
2754 		return 0;
2755 	}
2756 
2757 	if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2758 		return -EOPNOTSUPP;
2759 
2760 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2761 	if (!tmp)
2762 		return -ENOMEM;
2763 
2764 	if (prog) {
2765 		prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2766 		if (IS_ERR(prog)) {
2767 			err = PTR_ERR(prog);
2768 			goto out;
2769 		}
2770 	}
2771 
2772 	mutex_lock(&mdev->state_lock);
2773 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2774 	new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2775 
2776 	if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2777 		tx_changed = 1;
2778 		new_prof.tx_ring_num[TX] =
2779 			MAX_TX_RINGS - ALIGN(xdp_ring_num, MLX4_EN_NUM_UP);
2780 		en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2781 	}
2782 
2783 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2784 	if (err) {
2785 		if (prog)
2786 			bpf_prog_sub(prog, priv->rx_ring_num - 1);
2787 		goto unlock_out;
2788 	}
2789 
2790 	if (priv->port_up) {
2791 		port_up = 1;
2792 		mlx4_en_stop_port(dev, 1);
2793 	}
2794 
2795 	mlx4_en_safe_replace_resources(priv, tmp);
2796 	if (tx_changed)
2797 		netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2798 
2799 	for (i = 0; i < priv->rx_ring_num; i++) {
2800 		old_prog = rcu_dereference_protected(
2801 					priv->rx_ring[i]->xdp_prog,
2802 					lockdep_is_held(&mdev->state_lock));
2803 		rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2804 		if (old_prog)
2805 			bpf_prog_put(old_prog);
2806 	}
2807 
2808 	if (port_up) {
2809 		err = mlx4_en_start_port(dev);
2810 		if (err) {
2811 			en_err(priv, "Failed starting port %d for XDP change\n",
2812 			       priv->port);
2813 			queue_work(mdev->workqueue, &priv->watchdog_task);
2814 		}
2815 	}
2816 
2817 unlock_out:
2818 	mutex_unlock(&mdev->state_lock);
2819 out:
2820 	kfree(tmp);
2821 	return err;
2822 }
2823 
2824 static bool mlx4_xdp_attached(struct net_device *dev)
2825 {
2826 	struct mlx4_en_priv *priv = netdev_priv(dev);
2827 
2828 	return !!priv->tx_ring_num[TX_XDP];
2829 }
2830 
2831 static int mlx4_xdp(struct net_device *dev, struct netdev_xdp *xdp)
2832 {
2833 	switch (xdp->command) {
2834 	case XDP_SETUP_PROG:
2835 		return mlx4_xdp_set(dev, xdp->prog);
2836 	case XDP_QUERY_PROG:
2837 		xdp->prog_attached = mlx4_xdp_attached(dev);
2838 		return 0;
2839 	default:
2840 		return -EINVAL;
2841 	}
2842 }
2843 
2844 static const struct net_device_ops mlx4_netdev_ops = {
2845 	.ndo_open		= mlx4_en_open,
2846 	.ndo_stop		= mlx4_en_close,
2847 	.ndo_start_xmit		= mlx4_en_xmit,
2848 	.ndo_select_queue	= mlx4_en_select_queue,
2849 	.ndo_get_stats64	= mlx4_en_get_stats64,
2850 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2851 	.ndo_set_mac_address	= mlx4_en_set_mac,
2852 	.ndo_validate_addr	= eth_validate_addr,
2853 	.ndo_change_mtu		= mlx4_en_change_mtu,
2854 	.ndo_do_ioctl		= mlx4_en_ioctl,
2855 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2856 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2857 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2858 #ifdef CONFIG_NET_POLL_CONTROLLER
2859 	.ndo_poll_controller	= mlx4_en_netpoll,
2860 #endif
2861 	.ndo_set_features	= mlx4_en_set_features,
2862 	.ndo_fix_features	= mlx4_en_fix_features,
2863 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2864 #ifdef CONFIG_RFS_ACCEL
2865 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2866 #endif
2867 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2868 	.ndo_udp_tunnel_add	= mlx4_en_add_vxlan_port,
2869 	.ndo_udp_tunnel_del	= mlx4_en_del_vxlan_port,
2870 	.ndo_features_check	= mlx4_en_features_check,
2871 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2872 	.ndo_xdp		= mlx4_xdp,
2873 };
2874 
2875 static const struct net_device_ops mlx4_netdev_ops_master = {
2876 	.ndo_open		= mlx4_en_open,
2877 	.ndo_stop		= mlx4_en_close,
2878 	.ndo_start_xmit		= mlx4_en_xmit,
2879 	.ndo_select_queue	= mlx4_en_select_queue,
2880 	.ndo_get_stats64	= mlx4_en_get_stats64,
2881 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2882 	.ndo_set_mac_address	= mlx4_en_set_mac,
2883 	.ndo_validate_addr	= eth_validate_addr,
2884 	.ndo_change_mtu		= mlx4_en_change_mtu,
2885 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2886 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2887 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2888 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2889 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2890 	.ndo_set_vf_rate	= mlx4_en_set_vf_rate,
2891 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2892 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2893 	.ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2894 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2895 #ifdef CONFIG_NET_POLL_CONTROLLER
2896 	.ndo_poll_controller	= mlx4_en_netpoll,
2897 #endif
2898 	.ndo_set_features	= mlx4_en_set_features,
2899 	.ndo_fix_features	= mlx4_en_fix_features,
2900 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2901 #ifdef CONFIG_RFS_ACCEL
2902 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2903 #endif
2904 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2905 	.ndo_udp_tunnel_add	= mlx4_en_add_vxlan_port,
2906 	.ndo_udp_tunnel_del	= mlx4_en_del_vxlan_port,
2907 	.ndo_features_check	= mlx4_en_features_check,
2908 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2909 	.ndo_xdp		= mlx4_xdp,
2910 };
2911 
2912 struct mlx4_en_bond {
2913 	struct work_struct work;
2914 	struct mlx4_en_priv *priv;
2915 	int is_bonded;
2916 	struct mlx4_port_map port_map;
2917 };
2918 
2919 static void mlx4_en_bond_work(struct work_struct *work)
2920 {
2921 	struct mlx4_en_bond *bond = container_of(work,
2922 						     struct mlx4_en_bond,
2923 						     work);
2924 	int err = 0;
2925 	struct mlx4_dev *dev = bond->priv->mdev->dev;
2926 
2927 	if (bond->is_bonded) {
2928 		if (!mlx4_is_bonded(dev)) {
2929 			err = mlx4_bond(dev);
2930 			if (err)
2931 				en_err(bond->priv, "Fail to bond device\n");
2932 		}
2933 		if (!err) {
2934 			err = mlx4_port_map_set(dev, &bond->port_map);
2935 			if (err)
2936 				en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2937 				       bond->port_map.port1,
2938 				       bond->port_map.port2,
2939 				       err);
2940 		}
2941 	} else if (mlx4_is_bonded(dev)) {
2942 		err = mlx4_unbond(dev);
2943 		if (err)
2944 			en_err(bond->priv, "Fail to unbond device\n");
2945 	}
2946 	dev_put(bond->priv->dev);
2947 	kfree(bond);
2948 }
2949 
2950 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2951 				   u8 v2p_p1, u8 v2p_p2)
2952 {
2953 	struct mlx4_en_bond *bond = NULL;
2954 
2955 	bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2956 	if (!bond)
2957 		return -ENOMEM;
2958 
2959 	INIT_WORK(&bond->work, mlx4_en_bond_work);
2960 	bond->priv = priv;
2961 	bond->is_bonded = is_bonded;
2962 	bond->port_map.port1 = v2p_p1;
2963 	bond->port_map.port2 = v2p_p2;
2964 	dev_hold(priv->dev);
2965 	queue_work(priv->mdev->workqueue, &bond->work);
2966 	return 0;
2967 }
2968 
2969 int mlx4_en_netdev_event(struct notifier_block *this,
2970 			 unsigned long event, void *ptr)
2971 {
2972 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2973 	u8 port = 0;
2974 	struct mlx4_en_dev *mdev;
2975 	struct mlx4_dev *dev;
2976 	int i, num_eth_ports = 0;
2977 	bool do_bond = true;
2978 	struct mlx4_en_priv *priv;
2979 	u8 v2p_port1 = 0;
2980 	u8 v2p_port2 = 0;
2981 
2982 	if (!net_eq(dev_net(ndev), &init_net))
2983 		return NOTIFY_DONE;
2984 
2985 	mdev = container_of(this, struct mlx4_en_dev, nb);
2986 	dev = mdev->dev;
2987 
2988 	/* Go into this mode only when two network devices set on two ports
2989 	 * of the same mlx4 device are slaves of the same bonding master
2990 	 */
2991 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2992 		++num_eth_ports;
2993 		if (!port && (mdev->pndev[i] == ndev))
2994 			port = i;
2995 		mdev->upper[i] = mdev->pndev[i] ?
2996 			netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2997 		/* condition not met: network device is a slave */
2998 		if (!mdev->upper[i])
2999 			do_bond = false;
3000 		if (num_eth_ports < 2)
3001 			continue;
3002 		/* condition not met: same master */
3003 		if (mdev->upper[i] != mdev->upper[i-1])
3004 			do_bond = false;
3005 	}
3006 	/* condition not met: 2 salves */
3007 	do_bond = (num_eth_ports ==  2) ? do_bond : false;
3008 
3009 	/* handle only events that come with enough info */
3010 	if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
3011 		return NOTIFY_DONE;
3012 
3013 	priv = netdev_priv(ndev);
3014 	if (do_bond) {
3015 		struct netdev_notifier_bonding_info *notifier_info = ptr;
3016 		struct netdev_bonding_info *bonding_info =
3017 			&notifier_info->bonding_info;
3018 
3019 		/* required mode 1, 2 or 4 */
3020 		if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3021 		    (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
3022 		    (bonding_info->master.bond_mode != BOND_MODE_8023AD))
3023 			do_bond = false;
3024 
3025 		/* require exactly 2 slaves */
3026 		if (bonding_info->master.num_slaves != 2)
3027 			do_bond = false;
3028 
3029 		/* calc v2p */
3030 		if (do_bond) {
3031 			if (bonding_info->master.bond_mode ==
3032 			    BOND_MODE_ACTIVEBACKUP) {
3033 				/* in active-backup mode virtual ports are
3034 				 * mapped to the physical port of the active
3035 				 * slave */
3036 				if (bonding_info->slave.state ==
3037 				    BOND_STATE_BACKUP) {
3038 					if (port == 1) {
3039 						v2p_port1 = 2;
3040 						v2p_port2 = 2;
3041 					} else {
3042 						v2p_port1 = 1;
3043 						v2p_port2 = 1;
3044 					}
3045 				} else { /* BOND_STATE_ACTIVE */
3046 					if (port == 1) {
3047 						v2p_port1 = 1;
3048 						v2p_port2 = 1;
3049 					} else {
3050 						v2p_port1 = 2;
3051 						v2p_port2 = 2;
3052 					}
3053 				}
3054 			} else { /* Active-Active */
3055 				/* in active-active mode a virtual port is
3056 				 * mapped to the native physical port if and only
3057 				 * if the physical port is up */
3058 				__s8 link = bonding_info->slave.link;
3059 
3060 				if (port == 1)
3061 					v2p_port2 = 2;
3062 				else
3063 					v2p_port1 = 1;
3064 				if ((link == BOND_LINK_UP) ||
3065 				    (link == BOND_LINK_FAIL)) {
3066 					if (port == 1)
3067 						v2p_port1 = 1;
3068 					else
3069 						v2p_port2 = 2;
3070 				} else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3071 					if (port == 1)
3072 						v2p_port1 = 2;
3073 					else
3074 						v2p_port2 = 1;
3075 				}
3076 			}
3077 		}
3078 	}
3079 
3080 	mlx4_en_queue_bond_work(priv, do_bond,
3081 				v2p_port1, v2p_port2);
3082 
3083 	return NOTIFY_DONE;
3084 }
3085 
3086 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3087 				     struct mlx4_en_stats_bitmap *stats_bitmap,
3088 				     u8 rx_ppp, u8 rx_pause,
3089 				     u8 tx_ppp, u8 tx_pause)
3090 {
3091 	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3092 
3093 	if (!mlx4_is_slave(dev) &&
3094 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3095 		mutex_lock(&stats_bitmap->mutex);
3096 		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3097 
3098 		if (rx_ppp)
3099 			bitmap_set(stats_bitmap->bitmap, last_i,
3100 				   NUM_FLOW_PRIORITY_STATS_RX);
3101 		last_i += NUM_FLOW_PRIORITY_STATS_RX;
3102 
3103 		if (rx_pause && !(rx_ppp))
3104 			bitmap_set(stats_bitmap->bitmap, last_i,
3105 				   NUM_FLOW_STATS_RX);
3106 		last_i += NUM_FLOW_STATS_RX;
3107 
3108 		if (tx_ppp)
3109 			bitmap_set(stats_bitmap->bitmap, last_i,
3110 				   NUM_FLOW_PRIORITY_STATS_TX);
3111 		last_i += NUM_FLOW_PRIORITY_STATS_TX;
3112 
3113 		if (tx_pause && !(tx_ppp))
3114 			bitmap_set(stats_bitmap->bitmap, last_i,
3115 				   NUM_FLOW_STATS_TX);
3116 		last_i += NUM_FLOW_STATS_TX;
3117 
3118 		mutex_unlock(&stats_bitmap->mutex);
3119 	}
3120 }
3121 
3122 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3123 			      struct mlx4_en_stats_bitmap *stats_bitmap,
3124 			      u8 rx_ppp, u8 rx_pause,
3125 			      u8 tx_ppp, u8 tx_pause)
3126 {
3127 	int last_i = 0;
3128 
3129 	mutex_init(&stats_bitmap->mutex);
3130 	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3131 
3132 	if (mlx4_is_slave(dev)) {
3133 		bitmap_set(stats_bitmap->bitmap, last_i +
3134 					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3135 		bitmap_set(stats_bitmap->bitmap, last_i +
3136 					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3137 		bitmap_set(stats_bitmap->bitmap, last_i +
3138 					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3139 		bitmap_set(stats_bitmap->bitmap, last_i +
3140 					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3141 		bitmap_set(stats_bitmap->bitmap, last_i +
3142 					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3143 		bitmap_set(stats_bitmap->bitmap, last_i +
3144 					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3145 	} else {
3146 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3147 	}
3148 	last_i += NUM_MAIN_STATS;
3149 
3150 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3151 	last_i += NUM_PORT_STATS;
3152 
3153 	if (mlx4_is_master(dev))
3154 		bitmap_set(stats_bitmap->bitmap, last_i,
3155 			   NUM_PF_STATS);
3156 	last_i += NUM_PF_STATS;
3157 
3158 	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3159 					rx_ppp, rx_pause,
3160 					tx_ppp, tx_pause);
3161 	last_i += NUM_FLOW_STATS;
3162 
3163 	if (!mlx4_is_slave(dev))
3164 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3165 	last_i += NUM_PKT_STATS;
3166 
3167 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3168 	last_i += NUM_XDP_STATS;
3169 }
3170 
3171 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3172 			struct mlx4_en_port_profile *prof)
3173 {
3174 	struct net_device *dev;
3175 	struct mlx4_en_priv *priv;
3176 	int i, t;
3177 	int err;
3178 
3179 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3180 				 MAX_TX_RINGS, MAX_RX_RINGS);
3181 	if (dev == NULL)
3182 		return -ENOMEM;
3183 
3184 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3185 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3186 
3187 	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3188 	dev->dev_port = port - 1;
3189 
3190 	/*
3191 	 * Initialize driver private data
3192 	 */
3193 
3194 	priv = netdev_priv(dev);
3195 	memset(priv, 0, sizeof(struct mlx4_en_priv));
3196 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3197 	spin_lock_init(&priv->stats_lock);
3198 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3199 	INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
3200 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3201 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3202 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3203 	INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
3204 	INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
3205 #ifdef CONFIG_RFS_ACCEL
3206 	INIT_LIST_HEAD(&priv->filters);
3207 	spin_lock_init(&priv->filters_lock);
3208 #endif
3209 
3210 	priv->dev = dev;
3211 	priv->mdev = mdev;
3212 	priv->ddev = &mdev->pdev->dev;
3213 	priv->prof = prof;
3214 	priv->port = port;
3215 	priv->port_up = false;
3216 	priv->flags = prof->flags;
3217 	priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3218 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3219 			MLX4_WQE_CTRL_SOLICITED);
3220 	priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
3221 	priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3222 	netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3223 
3224 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3225 		priv->tx_ring_num[t] = prof->tx_ring_num[t];
3226 		if (!priv->tx_ring_num[t])
3227 			continue;
3228 
3229 		priv->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
3230 					   MAX_TX_RINGS, GFP_KERNEL);
3231 		if (!priv->tx_ring[t]) {
3232 			err = -ENOMEM;
3233 			goto err_free_tx;
3234 		}
3235 		priv->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
3236 					 MAX_TX_RINGS, GFP_KERNEL);
3237 		if (!priv->tx_cq[t]) {
3238 			kfree(priv->tx_ring[t]);
3239 			err = -ENOMEM;
3240 			goto out;
3241 		}
3242 	}
3243 	priv->rx_ring_num = prof->rx_ring_num;
3244 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3245 	priv->cqe_size = mdev->dev->caps.cqe_size;
3246 	priv->mac_index = -1;
3247 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
3248 #ifdef CONFIG_MLX4_EN_DCB
3249 	if (!mlx4_is_slave(priv->mdev->dev)) {
3250 		priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3251 			DCB_CAP_DCBX_VER_IEEE;
3252 		priv->flags |= MLX4_EN_DCB_ENABLED;
3253 		priv->cee_config.pfc_state = false;
3254 
3255 		for (i = 0; i < MLX4_EN_NUM_UP; i++)
3256 			priv->cee_config.dcb_pfc[i] = pfc_disabled;
3257 
3258 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3259 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3260 		} else {
3261 			en_info(priv, "enabling only PFC DCB ops\n");
3262 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3263 		}
3264 	}
3265 #endif
3266 
3267 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3268 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
3269 
3270 	/* Query for default mac and max mtu */
3271 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3272 
3273 	if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3274 	    MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3275 		priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3276 
3277 	/* Set default MAC */
3278 	dev->addr_len = ETH_ALEN;
3279 	mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3280 	if (!is_valid_ether_addr(dev->dev_addr)) {
3281 		en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
3282 		       priv->port, dev->dev_addr);
3283 		err = -EINVAL;
3284 		goto out;
3285 	} else if (mlx4_is_slave(priv->mdev->dev) &&
3286 		   (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3287 		/* Random MAC was assigned in mlx4_slave_cap
3288 		 * in mlx4_core module
3289 		 */
3290 		dev->addr_assign_type |= NET_ADDR_RANDOM;
3291 		en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3292 	}
3293 
3294 	memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3295 
3296 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3297 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3298 	err = mlx4_en_alloc_resources(priv);
3299 	if (err)
3300 		goto out;
3301 
3302 	/* Initialize time stamping config */
3303 	priv->hwtstamp_config.flags = 0;
3304 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3305 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3306 
3307 	/* Allocate page for receive rings */
3308 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3309 				MLX4_EN_PAGE_SIZE);
3310 	if (err) {
3311 		en_err(priv, "Failed to allocate page for rx qps\n");
3312 		goto out;
3313 	}
3314 	priv->allocated = 1;
3315 
3316 	/*
3317 	 * Initialize netdev entry points
3318 	 */
3319 	if (mlx4_is_master(priv->mdev->dev))
3320 		dev->netdev_ops = &mlx4_netdev_ops_master;
3321 	else
3322 		dev->netdev_ops = &mlx4_netdev_ops;
3323 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3324 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3325 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3326 
3327 	dev->ethtool_ops = &mlx4_en_ethtool_ops;
3328 
3329 	/*
3330 	 * Set driver features
3331 	 */
3332 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3333 	if (mdev->LSO_support)
3334 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3335 
3336 	dev->vlan_features = dev->hw_features;
3337 
3338 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3339 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3340 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3341 			NETIF_F_HW_VLAN_CTAG_FILTER;
3342 	dev->hw_features |= NETIF_F_LOOPBACK |
3343 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3344 
3345 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3346 		dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3347 			NETIF_F_HW_VLAN_STAG_FILTER;
3348 		dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3349 	}
3350 
3351 	if (mlx4_is_slave(mdev->dev)) {
3352 		bool vlan_offload_disabled;
3353 		int phv;
3354 
3355 		err = get_phv_bit(mdev->dev, port, &phv);
3356 		if (!err && phv) {
3357 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3358 			priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3359 		}
3360 		err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3361 							&vlan_offload_disabled);
3362 		if (!err && vlan_offload_disabled) {
3363 			dev->hw_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 			dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3368 					   NETIF_F_HW_VLAN_CTAG_RX |
3369 					   NETIF_F_HW_VLAN_STAG_TX |
3370 					   NETIF_F_HW_VLAN_STAG_RX);
3371 		}
3372 	} else {
3373 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3374 		    !(mdev->dev->caps.flags2 &
3375 		      MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3376 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3377 	}
3378 
3379 	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3380 		dev->hw_features |= NETIF_F_RXFCS;
3381 
3382 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3383 		dev->hw_features |= NETIF_F_RXALL;
3384 
3385 	if (mdev->dev->caps.steering_mode ==
3386 	    MLX4_STEERING_MODE_DEVICE_MANAGED &&
3387 	    mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3388 		dev->hw_features |= NETIF_F_NTUPLE;
3389 
3390 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3391 		dev->priv_flags |= IFF_UNICAST_FLT;
3392 
3393 	/* Setting a default hash function value */
3394 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3395 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3396 	} else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3397 		priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3398 	} else {
3399 		en_warn(priv,
3400 			"No RSS hash capabilities exposed, using Toeplitz\n");
3401 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3402 	}
3403 
3404 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3405 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3406 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3407 				    NETIF_F_GSO_PARTIAL;
3408 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3409 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3410 				    NETIF_F_GSO_PARTIAL;
3411 		dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3412 	}
3413 
3414 	/* MTU range: 46 - hw-specific max */
3415 	dev->min_mtu = MLX4_EN_MIN_MTU;
3416 	dev->max_mtu = priv->max_mtu;
3417 
3418 	mdev->pndev[port] = dev;
3419 	mdev->upper[port] = NULL;
3420 
3421 	netif_carrier_off(dev);
3422 	mlx4_en_set_default_moderation(priv);
3423 
3424 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3425 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3426 
3427 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3428 
3429 	/* Configure port */
3430 	mlx4_en_calc_rx_buf(dev);
3431 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3432 				    priv->rx_skb_size + ETH_FCS_LEN,
3433 				    prof->tx_pause, prof->tx_ppp,
3434 				    prof->rx_pause, prof->rx_ppp);
3435 	if (err) {
3436 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3437 		       priv->port, err);
3438 		goto out;
3439 	}
3440 
3441 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3442 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3443 		if (err) {
3444 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3445 			       err);
3446 			goto out;
3447 		}
3448 	}
3449 
3450 	/* Init port */
3451 	en_warn(priv, "Initializing port\n");
3452 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
3453 	if (err) {
3454 		en_err(priv, "Failed Initializing port\n");
3455 		goto out;
3456 	}
3457 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3458 
3459 	/* Initialize time stamp mechanism */
3460 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3461 		mlx4_en_init_timestamp(mdev);
3462 
3463 	queue_delayed_work(mdev->workqueue, &priv->service_task,
3464 			   SERVICE_TASK_DELAY);
3465 
3466 	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3467 				 mdev->profile.prof[priv->port].rx_ppp,
3468 				 mdev->profile.prof[priv->port].rx_pause,
3469 				 mdev->profile.prof[priv->port].tx_ppp,
3470 				 mdev->profile.prof[priv->port].tx_pause);
3471 
3472 	err = register_netdev(dev);
3473 	if (err) {
3474 		en_err(priv, "Netdev registration failed for port %d\n", port);
3475 		goto out;
3476 	}
3477 
3478 	priv->registered = 1;
3479 	devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3480 				  dev);
3481 
3482 	return 0;
3483 
3484 err_free_tx:
3485 	while (t--) {
3486 		kfree(priv->tx_ring[t]);
3487 		kfree(priv->tx_cq[t]);
3488 	}
3489 out:
3490 	mlx4_en_destroy_netdev(dev);
3491 	return err;
3492 }
3493 
3494 int mlx4_en_reset_config(struct net_device *dev,
3495 			 struct hwtstamp_config ts_config,
3496 			 netdev_features_t features)
3497 {
3498 	struct mlx4_en_priv *priv = netdev_priv(dev);
3499 	struct mlx4_en_dev *mdev = priv->mdev;
3500 	struct mlx4_en_port_profile new_prof;
3501 	struct mlx4_en_priv *tmp;
3502 	int port_up = 0;
3503 	int err = 0;
3504 
3505 	if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3506 	    priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3507 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3508 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3509 		return 0; /* Nothing to change */
3510 
3511 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3512 	    (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3513 	    (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3514 		en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3515 		return -EINVAL;
3516 	}
3517 
3518 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3519 	if (!tmp)
3520 		return -ENOMEM;
3521 
3522 	mutex_lock(&mdev->state_lock);
3523 
3524 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3525 	memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3526 
3527 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3528 	if (err)
3529 		goto out;
3530 
3531 	if (priv->port_up) {
3532 		port_up = 1;
3533 		mlx4_en_stop_port(dev, 1);
3534 	}
3535 
3536 	en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
3537 		ts_config.rx_filter,
3538 		!!(features & NETIF_F_HW_VLAN_CTAG_RX));
3539 
3540 	mlx4_en_safe_replace_resources(priv, tmp);
3541 
3542 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3543 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3544 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3545 		else
3546 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3547 	} else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3548 		/* RX time-stamping is OFF, update the RX vlan offload
3549 		 * to the latest wanted state
3550 		 */
3551 		if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3552 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3553 		else
3554 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3555 	}
3556 
3557 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3558 		if (features & NETIF_F_RXFCS)
3559 			dev->features |= NETIF_F_RXFCS;
3560 		else
3561 			dev->features &= ~NETIF_F_RXFCS;
3562 	}
3563 
3564 	/* RX vlan offload and RX time-stamping can't co-exist !
3565 	 * Regardless of the caller's choice,
3566 	 * Turn Off RX vlan offload in case of time-stamping is ON
3567 	 */
3568 	if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3569 		if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3570 			en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3571 		dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3572 	}
3573 
3574 	if (port_up) {
3575 		err = mlx4_en_start_port(dev);
3576 		if (err)
3577 			en_err(priv, "Failed starting port\n");
3578 	}
3579 
3580 out:
3581 	mutex_unlock(&mdev->state_lock);
3582 	kfree(tmp);
3583 	if (!err)
3584 		netdev_features_change(dev);
3585 	return err;
3586 }
3587