xref: /linux/drivers/net/ethernet/mellanox/mlx4/en_netdev.c (revision 32786fdc9506aeba98278c1844d4bfb766863832)
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 struct rtnl_link_stats64 *
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 	return stats;
1335 }
1336 
1337 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1338 {
1339 	struct mlx4_en_cq *cq;
1340 	int i, t;
1341 
1342 	/* If we haven't received a specific coalescing setting
1343 	 * (module param), we set the moderation parameters as follows:
1344 	 * - moder_cnt is set to the number of mtu sized packets to
1345 	 *   satisfy our coalescing target.
1346 	 * - moder_time is set to a fixed value.
1347 	 */
1348 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1349 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1350 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1351 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1352 	en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1353 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1354 
1355 	/* Setup cq moderation params */
1356 	for (i = 0; i < priv->rx_ring_num; i++) {
1357 		cq = priv->rx_cq[i];
1358 		cq->moder_cnt = priv->rx_frames;
1359 		cq->moder_time = priv->rx_usecs;
1360 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1361 		priv->last_moder_packets[i] = 0;
1362 		priv->last_moder_bytes[i] = 0;
1363 	}
1364 
1365 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1366 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1367 			cq = priv->tx_cq[t][i];
1368 			cq->moder_cnt = priv->tx_frames;
1369 			cq->moder_time = priv->tx_usecs;
1370 		}
1371 	}
1372 
1373 	/* Reset auto-moderation params */
1374 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1375 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1376 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1377 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1378 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1379 	priv->adaptive_rx_coal = 1;
1380 	priv->last_moder_jiffies = 0;
1381 	priv->last_moder_tx_packets = 0;
1382 }
1383 
1384 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1385 {
1386 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1387 	struct mlx4_en_cq *cq;
1388 	unsigned long packets;
1389 	unsigned long rate;
1390 	unsigned long avg_pkt_size;
1391 	unsigned long rx_packets;
1392 	unsigned long rx_bytes;
1393 	unsigned long rx_pkt_diff;
1394 	int moder_time;
1395 	int ring, err;
1396 
1397 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1398 		return;
1399 
1400 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1401 		rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1402 		rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1403 
1404 		rx_pkt_diff = ((unsigned long) (rx_packets -
1405 				priv->last_moder_packets[ring]));
1406 		packets = rx_pkt_diff;
1407 		rate = packets * HZ / period;
1408 		avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1409 				priv->last_moder_bytes[ring])) / packets : 0;
1410 
1411 		/* Apply auto-moderation only when packet rate
1412 		 * exceeds a rate that it matters */
1413 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1414 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1415 			if (rate < priv->pkt_rate_low)
1416 				moder_time = priv->rx_usecs_low;
1417 			else if (rate > priv->pkt_rate_high)
1418 				moder_time = priv->rx_usecs_high;
1419 			else
1420 				moder_time = (rate - priv->pkt_rate_low) *
1421 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1422 					(priv->pkt_rate_high - priv->pkt_rate_low) +
1423 					priv->rx_usecs_low;
1424 		} else {
1425 			moder_time = priv->rx_usecs_low;
1426 		}
1427 
1428 		if (moder_time != priv->last_moder_time[ring]) {
1429 			priv->last_moder_time[ring] = moder_time;
1430 			cq = priv->rx_cq[ring];
1431 			cq->moder_time = moder_time;
1432 			cq->moder_cnt = priv->rx_frames;
1433 			err = mlx4_en_set_cq_moder(priv, cq);
1434 			if (err)
1435 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1436 				       ring);
1437 		}
1438 		priv->last_moder_packets[ring] = rx_packets;
1439 		priv->last_moder_bytes[ring] = rx_bytes;
1440 	}
1441 
1442 	priv->last_moder_jiffies = jiffies;
1443 }
1444 
1445 static void mlx4_en_do_get_stats(struct work_struct *work)
1446 {
1447 	struct delayed_work *delay = to_delayed_work(work);
1448 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1449 						 stats_task);
1450 	struct mlx4_en_dev *mdev = priv->mdev;
1451 	int err;
1452 
1453 	mutex_lock(&mdev->state_lock);
1454 	if (mdev->device_up) {
1455 		if (priv->port_up) {
1456 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1457 			if (err)
1458 				en_dbg(HW, priv, "Could not update stats\n");
1459 
1460 			mlx4_en_auto_moderation(priv);
1461 		}
1462 
1463 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1464 	}
1465 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1466 		mlx4_en_do_set_mac(priv, priv->current_mac);
1467 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1468 	}
1469 	mutex_unlock(&mdev->state_lock);
1470 }
1471 
1472 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1473  * periodically
1474  */
1475 static void mlx4_en_service_task(struct work_struct *work)
1476 {
1477 	struct delayed_work *delay = to_delayed_work(work);
1478 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1479 						 service_task);
1480 	struct mlx4_en_dev *mdev = priv->mdev;
1481 
1482 	mutex_lock(&mdev->state_lock);
1483 	if (mdev->device_up) {
1484 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1485 			mlx4_en_ptp_overflow_check(mdev);
1486 
1487 		mlx4_en_recover_from_oom(priv);
1488 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1489 				   SERVICE_TASK_DELAY);
1490 	}
1491 	mutex_unlock(&mdev->state_lock);
1492 }
1493 
1494 static void mlx4_en_linkstate(struct work_struct *work)
1495 {
1496 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1497 						 linkstate_task);
1498 	struct mlx4_en_dev *mdev = priv->mdev;
1499 	int linkstate = priv->link_state;
1500 
1501 	mutex_lock(&mdev->state_lock);
1502 	/* If observable port state changed set carrier state and
1503 	 * report to system log */
1504 	if (priv->last_link_state != linkstate) {
1505 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1506 			en_info(priv, "Link Down\n");
1507 			netif_carrier_off(priv->dev);
1508 		} else {
1509 			en_info(priv, "Link Up\n");
1510 			netif_carrier_on(priv->dev);
1511 		}
1512 	}
1513 	priv->last_link_state = linkstate;
1514 	mutex_unlock(&mdev->state_lock);
1515 }
1516 
1517 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1518 {
1519 	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1520 	int numa_node = priv->mdev->dev->numa_node;
1521 
1522 	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1523 		return -ENOMEM;
1524 
1525 	cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1526 			ring->affinity_mask);
1527 	return 0;
1528 }
1529 
1530 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1531 {
1532 	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1533 }
1534 
1535 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1536 				      int tx_ring_idx)
1537 {
1538 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1539 	int rr_index = tx_ring_idx;
1540 
1541 	tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1542 	tx_ring->recycle_ring = priv->rx_ring[rr_index];
1543 	en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1544 	       TX_XDP, tx_ring_idx, rr_index);
1545 }
1546 
1547 int mlx4_en_start_port(struct net_device *dev)
1548 {
1549 	struct mlx4_en_priv *priv = netdev_priv(dev);
1550 	struct mlx4_en_dev *mdev = priv->mdev;
1551 	struct mlx4_en_cq *cq;
1552 	struct mlx4_en_tx_ring *tx_ring;
1553 	int rx_index = 0;
1554 	int err = 0;
1555 	int i, t;
1556 	int j;
1557 	u8 mc_list[16] = {0};
1558 
1559 	if (priv->port_up) {
1560 		en_dbg(DRV, priv, "start port called while port already up\n");
1561 		return 0;
1562 	}
1563 
1564 	INIT_LIST_HEAD(&priv->mc_list);
1565 	INIT_LIST_HEAD(&priv->curr_list);
1566 	INIT_LIST_HEAD(&priv->ethtool_list);
1567 	memset(&priv->ethtool_rules[0], 0,
1568 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1569 
1570 	/* Calculate Rx buf size */
1571 	dev->mtu = min(dev->mtu, priv->max_mtu);
1572 	mlx4_en_calc_rx_buf(dev);
1573 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1574 
1575 	/* Configure rx cq's and rings */
1576 	err = mlx4_en_activate_rx_rings(priv);
1577 	if (err) {
1578 		en_err(priv, "Failed to activate RX rings\n");
1579 		return err;
1580 	}
1581 	for (i = 0; i < priv->rx_ring_num; i++) {
1582 		cq = priv->rx_cq[i];
1583 
1584 		err = mlx4_en_init_affinity_hint(priv, i);
1585 		if (err) {
1586 			en_err(priv, "Failed preparing IRQ affinity hint\n");
1587 			goto cq_err;
1588 		}
1589 
1590 		err = mlx4_en_activate_cq(priv, cq, i);
1591 		if (err) {
1592 			en_err(priv, "Failed activating Rx CQ\n");
1593 			mlx4_en_free_affinity_hint(priv, i);
1594 			goto cq_err;
1595 		}
1596 
1597 		for (j = 0; j < cq->size; j++) {
1598 			struct mlx4_cqe *cqe = NULL;
1599 
1600 			cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1601 			      priv->cqe_factor;
1602 			cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1603 		}
1604 
1605 		err = mlx4_en_set_cq_moder(priv, cq);
1606 		if (err) {
1607 			en_err(priv, "Failed setting cq moderation parameters\n");
1608 			mlx4_en_deactivate_cq(priv, cq);
1609 			mlx4_en_free_affinity_hint(priv, i);
1610 			goto cq_err;
1611 		}
1612 		mlx4_en_arm_cq(priv, cq);
1613 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1614 		++rx_index;
1615 	}
1616 
1617 	/* Set qp number */
1618 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1619 	err = mlx4_en_get_qp(priv);
1620 	if (err) {
1621 		en_err(priv, "Failed getting eth qp\n");
1622 		goto cq_err;
1623 	}
1624 	mdev->mac_removed[priv->port] = 0;
1625 
1626 	priv->counter_index =
1627 			mlx4_get_default_counter_index(mdev->dev, priv->port);
1628 
1629 	err = mlx4_en_config_rss_steer(priv);
1630 	if (err) {
1631 		en_err(priv, "Failed configuring rss steering\n");
1632 		goto mac_err;
1633 	}
1634 
1635 	err = mlx4_en_create_drop_qp(priv);
1636 	if (err)
1637 		goto rss_err;
1638 
1639 	/* Configure tx cq's and rings */
1640 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1641 		u8 num_tx_rings_p_up = t == TX ? priv->num_tx_rings_p_up : 1;
1642 
1643 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1644 			/* Configure cq */
1645 			cq = priv->tx_cq[t][i];
1646 			err = mlx4_en_activate_cq(priv, cq, i);
1647 			if (err) {
1648 				en_err(priv, "Failed allocating Tx CQ\n");
1649 				goto tx_err;
1650 			}
1651 			err = mlx4_en_set_cq_moder(priv, cq);
1652 			if (err) {
1653 				en_err(priv, "Failed setting cq moderation parameters\n");
1654 				mlx4_en_deactivate_cq(priv, cq);
1655 				goto tx_err;
1656 			}
1657 			en_dbg(DRV, priv,
1658 			       "Resetting index of collapsed CQ:%d to -1\n", i);
1659 			cq->buf->wqe_index = cpu_to_be16(0xffff);
1660 
1661 			/* Configure ring */
1662 			tx_ring = priv->tx_ring[t][i];
1663 			err = mlx4_en_activate_tx_ring(priv, tx_ring,
1664 						       cq->mcq.cqn,
1665 						       i / num_tx_rings_p_up);
1666 			if (err) {
1667 				en_err(priv, "Failed allocating Tx ring\n");
1668 				mlx4_en_deactivate_cq(priv, cq);
1669 				goto tx_err;
1670 			}
1671 			if (t != TX_XDP) {
1672 				tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1673 				tx_ring->recycle_ring = NULL;
1674 			} else {
1675 				mlx4_en_init_recycle_ring(priv, i);
1676 			}
1677 
1678 			/* Arm CQ for TX completions */
1679 			mlx4_en_arm_cq(priv, cq);
1680 
1681 			/* Set initial ownership of all Tx TXBBs to SW (1) */
1682 			for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1683 				*((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1684 		}
1685 	}
1686 
1687 	/* Configure port */
1688 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1689 				    priv->rx_skb_size + ETH_FCS_LEN,
1690 				    priv->prof->tx_pause,
1691 				    priv->prof->tx_ppp,
1692 				    priv->prof->rx_pause,
1693 				    priv->prof->rx_ppp);
1694 	if (err) {
1695 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1696 		       priv->port, err);
1697 		goto tx_err;
1698 	}
1699 	/* Set default qp number */
1700 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1701 	if (err) {
1702 		en_err(priv, "Failed setting default qp numbers\n");
1703 		goto tx_err;
1704 	}
1705 
1706 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1707 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1708 		if (err) {
1709 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1710 			       err);
1711 			goto tx_err;
1712 		}
1713 	}
1714 
1715 	/* Init port */
1716 	en_dbg(HW, priv, "Initializing port\n");
1717 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1718 	if (err) {
1719 		en_err(priv, "Failed Initializing port\n");
1720 		goto tx_err;
1721 	}
1722 
1723 	/* Set Unicast and VXLAN steering rules */
1724 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1725 	    mlx4_en_set_rss_steer_rules(priv))
1726 		mlx4_warn(mdev, "Failed setting steering rules\n");
1727 
1728 	/* Attach rx QP to bradcast address */
1729 	eth_broadcast_addr(&mc_list[10]);
1730 	mc_list[5] = priv->port; /* needed for B0 steering support */
1731 	if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1732 				  priv->port, 0, MLX4_PROT_ETH,
1733 				  &priv->broadcast_id))
1734 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1735 
1736 	/* Must redo promiscuous mode setup. */
1737 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1738 
1739 	/* Schedule multicast task to populate multicast list */
1740 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1741 
1742 	if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1743 		udp_tunnel_get_rx_info(dev);
1744 
1745 	priv->port_up = true;
1746 
1747 	/* Process all completions if exist to prevent
1748 	 * the queues freezing if they are full
1749 	 */
1750 	for (i = 0; i < priv->rx_ring_num; i++)
1751 		napi_schedule(&priv->rx_cq[i]->napi);
1752 
1753 	netif_tx_start_all_queues(dev);
1754 	netif_device_attach(dev);
1755 
1756 	return 0;
1757 
1758 tx_err:
1759 	if (t == MLX4_EN_NUM_TX_TYPES) {
1760 		t--;
1761 		i = priv->tx_ring_num[t];
1762 	}
1763 	while (t >= 0) {
1764 		while (i--) {
1765 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1766 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1767 		}
1768 		if (!t--)
1769 			break;
1770 		i = priv->tx_ring_num[t];
1771 	}
1772 	mlx4_en_destroy_drop_qp(priv);
1773 rss_err:
1774 	mlx4_en_release_rss_steer(priv);
1775 mac_err:
1776 	mlx4_en_put_qp(priv);
1777 cq_err:
1778 	while (rx_index--) {
1779 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1780 		mlx4_en_free_affinity_hint(priv, rx_index);
1781 	}
1782 	for (i = 0; i < priv->rx_ring_num; i++)
1783 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1784 
1785 	return err; /* need to close devices */
1786 }
1787 
1788 
1789 void mlx4_en_stop_port(struct net_device *dev, int detach)
1790 {
1791 	struct mlx4_en_priv *priv = netdev_priv(dev);
1792 	struct mlx4_en_dev *mdev = priv->mdev;
1793 	struct mlx4_en_mc_list *mclist, *tmp;
1794 	struct ethtool_flow_id *flow, *tmp_flow;
1795 	int i, t;
1796 	u8 mc_list[16] = {0};
1797 
1798 	if (!priv->port_up) {
1799 		en_dbg(DRV, priv, "stop port called while port already down\n");
1800 		return;
1801 	}
1802 
1803 	/* close port*/
1804 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1805 
1806 	/* Synchronize with tx routine */
1807 	netif_tx_lock_bh(dev);
1808 	if (detach)
1809 		netif_device_detach(dev);
1810 	netif_tx_stop_all_queues(dev);
1811 	netif_tx_unlock_bh(dev);
1812 
1813 	netif_tx_disable(dev);
1814 
1815 	spin_lock_bh(&priv->stats_lock);
1816 	mlx4_en_fold_software_stats(dev);
1817 	/* Set port as not active */
1818 	priv->port_up = false;
1819 	spin_unlock_bh(&priv->stats_lock);
1820 
1821 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1822 
1823 	/* Promsicuous mode */
1824 	if (mdev->dev->caps.steering_mode ==
1825 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1826 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1827 				 MLX4_EN_FLAG_MC_PROMISC);
1828 		mlx4_flow_steer_promisc_remove(mdev->dev,
1829 					       priv->port,
1830 					       MLX4_FS_ALL_DEFAULT);
1831 		mlx4_flow_steer_promisc_remove(mdev->dev,
1832 					       priv->port,
1833 					       MLX4_FS_MC_DEFAULT);
1834 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1835 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1836 
1837 		/* Disable promiscouos mode */
1838 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1839 					    priv->port);
1840 
1841 		/* Disable Multicast promisc */
1842 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1843 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1844 						      priv->port);
1845 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1846 		}
1847 	}
1848 
1849 	/* Detach All multicasts */
1850 	eth_broadcast_addr(&mc_list[10]);
1851 	mc_list[5] = priv->port; /* needed for B0 steering support */
1852 	mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1853 			      MLX4_PROT_ETH, priv->broadcast_id);
1854 	list_for_each_entry(mclist, &priv->curr_list, list) {
1855 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1856 		mc_list[5] = priv->port;
1857 		mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1858 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1859 		if (mclist->tunnel_reg_id)
1860 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1861 	}
1862 	mlx4_en_clear_list(dev);
1863 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1864 		list_del(&mclist->list);
1865 		kfree(mclist);
1866 	}
1867 
1868 	/* Flush multicast filter */
1869 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1870 
1871 	/* Remove flow steering rules for the port*/
1872 	if (mdev->dev->caps.steering_mode ==
1873 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1874 		ASSERT_RTNL();
1875 		list_for_each_entry_safe(flow, tmp_flow,
1876 					 &priv->ethtool_list, list) {
1877 			mlx4_flow_detach(mdev->dev, flow->id);
1878 			list_del(&flow->list);
1879 		}
1880 	}
1881 
1882 	mlx4_en_destroy_drop_qp(priv);
1883 
1884 	/* Free TX Rings */
1885 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1886 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1887 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1888 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1889 		}
1890 	}
1891 	msleep(10);
1892 
1893 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1894 		for (i = 0; i < priv->tx_ring_num[t]; i++)
1895 			mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1896 
1897 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1898 		mlx4_en_delete_rss_steer_rules(priv);
1899 
1900 	/* Free RSS qps */
1901 	mlx4_en_release_rss_steer(priv);
1902 
1903 	/* Unregister Mac address for the port */
1904 	mlx4_en_put_qp(priv);
1905 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1906 		mdev->mac_removed[priv->port] = 1;
1907 
1908 	/* Free RX Rings */
1909 	for (i = 0; i < priv->rx_ring_num; i++) {
1910 		struct mlx4_en_cq *cq = priv->rx_cq[i];
1911 
1912 		napi_synchronize(&cq->napi);
1913 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1914 		mlx4_en_deactivate_cq(priv, cq);
1915 
1916 		mlx4_en_free_affinity_hint(priv, i);
1917 	}
1918 }
1919 
1920 static void mlx4_en_restart(struct work_struct *work)
1921 {
1922 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1923 						 watchdog_task);
1924 	struct mlx4_en_dev *mdev = priv->mdev;
1925 	struct net_device *dev = priv->dev;
1926 
1927 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1928 
1929 	rtnl_lock();
1930 	mutex_lock(&mdev->state_lock);
1931 	if (priv->port_up) {
1932 		mlx4_en_stop_port(dev, 1);
1933 		if (mlx4_en_start_port(dev))
1934 			en_err(priv, "Failed restarting port %d\n", priv->port);
1935 	}
1936 	mutex_unlock(&mdev->state_lock);
1937 	rtnl_unlock();
1938 }
1939 
1940 static void mlx4_en_clear_stats(struct net_device *dev)
1941 {
1942 	struct mlx4_en_priv *priv = netdev_priv(dev);
1943 	struct mlx4_en_dev *mdev = priv->mdev;
1944 	struct mlx4_en_tx_ring **tx_ring;
1945 	int i;
1946 
1947 	if (!mlx4_is_slave(mdev->dev))
1948 		if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1949 			en_dbg(HW, priv, "Failed dumping statistics\n");
1950 
1951 	memset(&priv->pstats, 0, sizeof(priv->pstats));
1952 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1953 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1954 	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
1955 	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
1956 	memset(&priv->rx_priority_flowstats, 0,
1957 	       sizeof(priv->rx_priority_flowstats));
1958 	memset(&priv->tx_priority_flowstats, 0,
1959 	       sizeof(priv->tx_priority_flowstats));
1960 	memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
1961 
1962 	tx_ring = priv->tx_ring[TX];
1963 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1964 		tx_ring[i]->bytes = 0;
1965 		tx_ring[i]->packets = 0;
1966 		tx_ring[i]->tx_csum = 0;
1967 		tx_ring[i]->tx_dropped = 0;
1968 		tx_ring[i]->queue_stopped = 0;
1969 		tx_ring[i]->wake_queue = 0;
1970 		tx_ring[i]->tso_packets = 0;
1971 		tx_ring[i]->xmit_more = 0;
1972 	}
1973 	for (i = 0; i < priv->rx_ring_num; i++) {
1974 		priv->rx_ring[i]->bytes = 0;
1975 		priv->rx_ring[i]->packets = 0;
1976 		priv->rx_ring[i]->csum_ok = 0;
1977 		priv->rx_ring[i]->csum_none = 0;
1978 		priv->rx_ring[i]->csum_complete = 0;
1979 	}
1980 }
1981 
1982 static int mlx4_en_open(struct net_device *dev)
1983 {
1984 	struct mlx4_en_priv *priv = netdev_priv(dev);
1985 	struct mlx4_en_dev *mdev = priv->mdev;
1986 	int err = 0;
1987 
1988 	mutex_lock(&mdev->state_lock);
1989 
1990 	if (!mdev->device_up) {
1991 		en_err(priv, "Cannot open - device down/disabled\n");
1992 		err = -EBUSY;
1993 		goto out;
1994 	}
1995 
1996 	/* Reset HW statistics and SW counters */
1997 	mlx4_en_clear_stats(dev);
1998 
1999 	err = mlx4_en_start_port(dev);
2000 	if (err)
2001 		en_err(priv, "Failed starting port:%d\n", priv->port);
2002 
2003 out:
2004 	mutex_unlock(&mdev->state_lock);
2005 	return err;
2006 }
2007 
2008 
2009 static int mlx4_en_close(struct net_device *dev)
2010 {
2011 	struct mlx4_en_priv *priv = netdev_priv(dev);
2012 	struct mlx4_en_dev *mdev = priv->mdev;
2013 
2014 	en_dbg(IFDOWN, priv, "Close port called\n");
2015 
2016 	mutex_lock(&mdev->state_lock);
2017 
2018 	mlx4_en_stop_port(dev, 0);
2019 	netif_carrier_off(dev);
2020 
2021 	mutex_unlock(&mdev->state_lock);
2022 	return 0;
2023 }
2024 
2025 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2026 {
2027 	int i, t;
2028 
2029 #ifdef CONFIG_RFS_ACCEL
2030 	priv->dev->rx_cpu_rmap = NULL;
2031 #endif
2032 
2033 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2034 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2035 			if (priv->tx_ring[t] && priv->tx_ring[t][i])
2036 				mlx4_en_destroy_tx_ring(priv,
2037 							&priv->tx_ring[t][i]);
2038 			if (priv->tx_cq[t] && priv->tx_cq[t][i])
2039 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2040 		}
2041 	}
2042 
2043 	for (i = 0; i < priv->rx_ring_num; i++) {
2044 		if (priv->rx_ring[i])
2045 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2046 				priv->prof->rx_ring_size, priv->stride);
2047 		if (priv->rx_cq[i])
2048 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2049 	}
2050 
2051 }
2052 
2053 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2054 {
2055 	struct mlx4_en_port_profile *prof = priv->prof;
2056 	int i, t;
2057 	int node;
2058 
2059 	/* Create tx Rings */
2060 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2061 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2062 			node = cpu_to_node(i % num_online_cpus());
2063 			if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2064 					      prof->tx_ring_size, i, t, node))
2065 				goto err;
2066 
2067 			if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2068 						   prof->tx_ring_size,
2069 						   TXBB_SIZE, node, i))
2070 				goto err;
2071 		}
2072 	}
2073 
2074 	/* Create rx Rings */
2075 	for (i = 0; i < priv->rx_ring_num; i++) {
2076 		node = cpu_to_node(i % num_online_cpus());
2077 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2078 				      prof->rx_ring_size, i, RX, node))
2079 			goto err;
2080 
2081 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2082 					   prof->rx_ring_size, priv->stride,
2083 					   node))
2084 			goto err;
2085 	}
2086 
2087 #ifdef CONFIG_RFS_ACCEL
2088 	priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2089 #endif
2090 
2091 	return 0;
2092 
2093 err:
2094 	en_err(priv, "Failed to allocate NIC resources\n");
2095 	for (i = 0; i < priv->rx_ring_num; i++) {
2096 		if (priv->rx_ring[i])
2097 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2098 						prof->rx_ring_size,
2099 						priv->stride);
2100 		if (priv->rx_cq[i])
2101 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2102 	}
2103 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2104 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2105 			if (priv->tx_ring[t][i])
2106 				mlx4_en_destroy_tx_ring(priv,
2107 							&priv->tx_ring[t][i]);
2108 			if (priv->tx_cq[t][i])
2109 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2110 		}
2111 	}
2112 	return -ENOMEM;
2113 }
2114 
2115 
2116 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2117 			     struct mlx4_en_priv *src,
2118 			     struct mlx4_en_port_profile *prof)
2119 {
2120 	int t;
2121 
2122 	memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2123 	       sizeof(dst->hwtstamp_config));
2124 	dst->num_tx_rings_p_up = src->mdev->profile.num_tx_rings_p_up;
2125 	dst->rx_ring_num = prof->rx_ring_num;
2126 	dst->flags = prof->flags;
2127 	dst->mdev = src->mdev;
2128 	dst->port = src->port;
2129 	dst->dev = src->dev;
2130 	dst->prof = prof;
2131 	dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2132 					 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2133 
2134 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2135 		dst->tx_ring_num[t] = prof->tx_ring_num[t];
2136 		if (!dst->tx_ring_num[t])
2137 			continue;
2138 
2139 		dst->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
2140 					  MAX_TX_RINGS, GFP_KERNEL);
2141 		if (!dst->tx_ring[t])
2142 			goto err_free_tx;
2143 
2144 		dst->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
2145 					MAX_TX_RINGS, GFP_KERNEL);
2146 		if (!dst->tx_cq[t]) {
2147 			kfree(dst->tx_ring[t]);
2148 			goto err_free_tx;
2149 		}
2150 	}
2151 
2152 	return 0;
2153 
2154 err_free_tx:
2155 	while (t--) {
2156 		kfree(dst->tx_ring[t]);
2157 		kfree(dst->tx_cq[t]);
2158 	}
2159 	return -ENOMEM;
2160 }
2161 
2162 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2163 				struct mlx4_en_priv *src)
2164 {
2165 	int t;
2166 	memcpy(dst->rx_ring, src->rx_ring,
2167 	       sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2168 	memcpy(dst->rx_cq, src->rx_cq,
2169 	       sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2170 	memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2171 	       sizeof(dst->hwtstamp_config));
2172 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2173 		dst->tx_ring_num[t] = src->tx_ring_num[t];
2174 		dst->tx_ring[t] = src->tx_ring[t];
2175 		dst->tx_cq[t] = src->tx_cq[t];
2176 	}
2177 	dst->rx_ring_num = src->rx_ring_num;
2178 	memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2179 }
2180 
2181 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2182 				struct mlx4_en_priv *tmp,
2183 				struct mlx4_en_port_profile *prof)
2184 {
2185 	int t;
2186 
2187 	mlx4_en_copy_priv(tmp, priv, prof);
2188 
2189 	if (mlx4_en_alloc_resources(tmp)) {
2190 		en_warn(priv,
2191 			"%s: Resource allocation failed, using previous configuration\n",
2192 			__func__);
2193 		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2194 			kfree(tmp->tx_ring[t]);
2195 			kfree(tmp->tx_cq[t]);
2196 		}
2197 		return -ENOMEM;
2198 	}
2199 	return 0;
2200 }
2201 
2202 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2203 				    struct mlx4_en_priv *tmp)
2204 {
2205 	mlx4_en_free_resources(priv);
2206 	mlx4_en_update_priv(priv, tmp);
2207 }
2208 
2209 void mlx4_en_destroy_netdev(struct net_device *dev)
2210 {
2211 	struct mlx4_en_priv *priv = netdev_priv(dev);
2212 	struct mlx4_en_dev *mdev = priv->mdev;
2213 	int t;
2214 
2215 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2216 
2217 	/* Unregister device - this will close the port if it was up */
2218 	if (priv->registered) {
2219 		devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2220 							      priv->port));
2221 		unregister_netdev(dev);
2222 	}
2223 
2224 	if (priv->allocated)
2225 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2226 
2227 	cancel_delayed_work(&priv->stats_task);
2228 	cancel_delayed_work(&priv->service_task);
2229 	/* flush any pending task for this netdev */
2230 	flush_workqueue(mdev->workqueue);
2231 
2232 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2233 		mlx4_en_remove_timestamp(mdev);
2234 
2235 	/* Detach the netdev so tasks would not attempt to access it */
2236 	mutex_lock(&mdev->state_lock);
2237 	mdev->pndev[priv->port] = NULL;
2238 	mdev->upper[priv->port] = NULL;
2239 
2240 #ifdef CONFIG_RFS_ACCEL
2241 	mlx4_en_cleanup_filters(priv);
2242 #endif
2243 
2244 	mlx4_en_free_resources(priv);
2245 	mutex_unlock(&mdev->state_lock);
2246 
2247 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2248 		kfree(priv->tx_ring[t]);
2249 		kfree(priv->tx_cq[t]);
2250 	}
2251 
2252 	free_netdev(dev);
2253 }
2254 
2255 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2256 {
2257 	struct mlx4_en_priv *priv = netdev_priv(dev);
2258 
2259 	if (mtu > MLX4_EN_MAX_XDP_MTU) {
2260 		en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2261 		       mtu, MLX4_EN_MAX_XDP_MTU);
2262 		return false;
2263 	}
2264 
2265 	return true;
2266 }
2267 
2268 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2269 {
2270 	struct mlx4_en_priv *priv = netdev_priv(dev);
2271 	struct mlx4_en_dev *mdev = priv->mdev;
2272 	int err = 0;
2273 
2274 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2275 		 dev->mtu, new_mtu);
2276 
2277 	if (priv->tx_ring_num[TX_XDP] &&
2278 	    !mlx4_en_check_xdp_mtu(dev, new_mtu))
2279 		return -ENOTSUPP;
2280 
2281 	dev->mtu = new_mtu;
2282 
2283 	if (netif_running(dev)) {
2284 		mutex_lock(&mdev->state_lock);
2285 		if (!mdev->device_up) {
2286 			/* NIC is probably restarting - let watchdog task reset
2287 			 * the port */
2288 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2289 		} else {
2290 			mlx4_en_stop_port(dev, 1);
2291 			err = mlx4_en_start_port(dev);
2292 			if (err) {
2293 				en_err(priv, "Failed restarting port:%d\n",
2294 					 priv->port);
2295 				queue_work(mdev->workqueue, &priv->watchdog_task);
2296 			}
2297 		}
2298 		mutex_unlock(&mdev->state_lock);
2299 	}
2300 	return 0;
2301 }
2302 
2303 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2304 {
2305 	struct mlx4_en_priv *priv = netdev_priv(dev);
2306 	struct mlx4_en_dev *mdev = priv->mdev;
2307 	struct hwtstamp_config config;
2308 
2309 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2310 		return -EFAULT;
2311 
2312 	/* reserved for future extensions */
2313 	if (config.flags)
2314 		return -EINVAL;
2315 
2316 	/* device doesn't support time stamping */
2317 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2318 		return -EINVAL;
2319 
2320 	/* TX HW timestamp */
2321 	switch (config.tx_type) {
2322 	case HWTSTAMP_TX_OFF:
2323 	case HWTSTAMP_TX_ON:
2324 		break;
2325 	default:
2326 		return -ERANGE;
2327 	}
2328 
2329 	/* RX HW timestamp */
2330 	switch (config.rx_filter) {
2331 	case HWTSTAMP_FILTER_NONE:
2332 		break;
2333 	case HWTSTAMP_FILTER_ALL:
2334 	case HWTSTAMP_FILTER_SOME:
2335 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2336 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2337 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2338 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2339 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2340 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2341 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2342 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2343 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2344 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2345 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2346 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2347 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2348 		break;
2349 	default:
2350 		return -ERANGE;
2351 	}
2352 
2353 	if (mlx4_en_reset_config(dev, config, dev->features)) {
2354 		config.tx_type = HWTSTAMP_TX_OFF;
2355 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2356 	}
2357 
2358 	return copy_to_user(ifr->ifr_data, &config,
2359 			    sizeof(config)) ? -EFAULT : 0;
2360 }
2361 
2362 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2363 {
2364 	struct mlx4_en_priv *priv = netdev_priv(dev);
2365 
2366 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2367 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2368 }
2369 
2370 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2371 {
2372 	switch (cmd) {
2373 	case SIOCSHWTSTAMP:
2374 		return mlx4_en_hwtstamp_set(dev, ifr);
2375 	case SIOCGHWTSTAMP:
2376 		return mlx4_en_hwtstamp_get(dev, ifr);
2377 	default:
2378 		return -EOPNOTSUPP;
2379 	}
2380 }
2381 
2382 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2383 					      netdev_features_t features)
2384 {
2385 	struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2386 	struct mlx4_en_dev *mdev = en_priv->mdev;
2387 
2388 	/* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2389 	 * enable/disable make sure S-TAG flag is always in same state as
2390 	 * C-TAG.
2391 	 */
2392 	if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2393 	    !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2394 		features |= NETIF_F_HW_VLAN_STAG_RX;
2395 	else
2396 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2397 
2398 	return features;
2399 }
2400 
2401 static int mlx4_en_set_features(struct net_device *netdev,
2402 		netdev_features_t features)
2403 {
2404 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2405 	bool reset = false;
2406 	int ret = 0;
2407 
2408 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2409 		en_info(priv, "Turn %s RX-FCS\n",
2410 			(features & NETIF_F_RXFCS) ? "ON" : "OFF");
2411 		reset = true;
2412 	}
2413 
2414 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2415 		u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2416 
2417 		en_info(priv, "Turn %s RX-ALL\n",
2418 			ignore_fcs_value ? "ON" : "OFF");
2419 		ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2420 					      priv->port, ignore_fcs_value);
2421 		if (ret)
2422 			return ret;
2423 	}
2424 
2425 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2426 		en_info(priv, "Turn %s RX vlan strip offload\n",
2427 			(features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2428 		reset = true;
2429 	}
2430 
2431 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2432 		en_info(priv, "Turn %s TX vlan strip offload\n",
2433 			(features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2434 
2435 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2436 		en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2437 			(features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2438 
2439 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2440 		en_info(priv, "Turn %s loopback\n",
2441 			(features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2442 		mlx4_en_update_loopback_state(netdev, features);
2443 	}
2444 
2445 	if (reset) {
2446 		ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2447 					   features);
2448 		if (ret)
2449 			return ret;
2450 	}
2451 
2452 	return 0;
2453 }
2454 
2455 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2456 {
2457 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2458 	struct mlx4_en_dev *mdev = en_priv->mdev;
2459 	u64 mac_u64 = mlx4_mac_to_u64(mac);
2460 
2461 	if (is_multicast_ether_addr(mac))
2462 		return -EINVAL;
2463 
2464 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2465 }
2466 
2467 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2468 			       __be16 vlan_proto)
2469 {
2470 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2471 	struct mlx4_en_dev *mdev = en_priv->mdev;
2472 
2473 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2474 				vlan_proto);
2475 }
2476 
2477 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2478 			       int max_tx_rate)
2479 {
2480 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2481 	struct mlx4_en_dev *mdev = en_priv->mdev;
2482 
2483 	return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2484 				max_tx_rate);
2485 }
2486 
2487 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2488 {
2489 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2490 	struct mlx4_en_dev *mdev = en_priv->mdev;
2491 
2492 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2493 }
2494 
2495 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2496 {
2497 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2498 	struct mlx4_en_dev *mdev = en_priv->mdev;
2499 
2500 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2501 }
2502 
2503 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2504 {
2505 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2506 	struct mlx4_en_dev *mdev = en_priv->mdev;
2507 
2508 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2509 }
2510 
2511 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2512 				struct ifla_vf_stats *vf_stats)
2513 {
2514 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2515 	struct mlx4_en_dev *mdev = en_priv->mdev;
2516 
2517 	return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2518 }
2519 
2520 #define PORT_ID_BYTE_LEN 8
2521 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2522 				    struct netdev_phys_item_id *ppid)
2523 {
2524 	struct mlx4_en_priv *priv = netdev_priv(dev);
2525 	struct mlx4_dev *mdev = priv->mdev->dev;
2526 	int i;
2527 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2528 
2529 	if (!phys_port_id)
2530 		return -EOPNOTSUPP;
2531 
2532 	ppid->id_len = sizeof(phys_port_id);
2533 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2534 		ppid->id[i] =  phys_port_id & 0xff;
2535 		phys_port_id >>= 8;
2536 	}
2537 	return 0;
2538 }
2539 
2540 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2541 {
2542 	int ret;
2543 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2544 						 vxlan_add_task);
2545 
2546 	ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2547 	if (ret)
2548 		goto out;
2549 
2550 	ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2551 				  VXLAN_STEER_BY_OUTER_MAC, 1);
2552 out:
2553 	if (ret) {
2554 		en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2555 		return;
2556 	}
2557 
2558 	/* set offloads */
2559 	priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2560 				      NETIF_F_RXCSUM |
2561 				      NETIF_F_TSO | NETIF_F_TSO6 |
2562 				      NETIF_F_GSO_UDP_TUNNEL |
2563 				      NETIF_F_GSO_UDP_TUNNEL_CSUM |
2564 				      NETIF_F_GSO_PARTIAL;
2565 }
2566 
2567 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2568 {
2569 	int ret;
2570 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2571 						 vxlan_del_task);
2572 	/* unset offloads */
2573 	priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2574 					NETIF_F_RXCSUM |
2575 					NETIF_F_TSO | NETIF_F_TSO6 |
2576 					NETIF_F_GSO_UDP_TUNNEL |
2577 					NETIF_F_GSO_UDP_TUNNEL_CSUM |
2578 					NETIF_F_GSO_PARTIAL);
2579 
2580 	ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2581 				  VXLAN_STEER_BY_OUTER_MAC, 0);
2582 	if (ret)
2583 		en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2584 
2585 	priv->vxlan_port = 0;
2586 }
2587 
2588 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2589 				   struct udp_tunnel_info *ti)
2590 {
2591 	struct mlx4_en_priv *priv = netdev_priv(dev);
2592 	__be16 port = ti->port;
2593 	__be16 current_port;
2594 
2595 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2596 		return;
2597 
2598 	if (ti->sa_family != AF_INET)
2599 		return;
2600 
2601 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2602 		return;
2603 
2604 	current_port = priv->vxlan_port;
2605 	if (current_port && current_port != port) {
2606 		en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2607 			ntohs(current_port), ntohs(port));
2608 		return;
2609 	}
2610 
2611 	priv->vxlan_port = port;
2612 	queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2613 }
2614 
2615 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2616 				   struct udp_tunnel_info *ti)
2617 {
2618 	struct mlx4_en_priv *priv = netdev_priv(dev);
2619 	__be16 port = ti->port;
2620 	__be16 current_port;
2621 
2622 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2623 		return;
2624 
2625 	if (ti->sa_family != AF_INET)
2626 		return;
2627 
2628 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2629 		return;
2630 
2631 	current_port = priv->vxlan_port;
2632 	if (current_port != port) {
2633 		en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2634 		return;
2635 	}
2636 
2637 	queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2638 }
2639 
2640 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2641 						struct net_device *dev,
2642 						netdev_features_t features)
2643 {
2644 	features = vlan_features_check(skb, features);
2645 	features = vxlan_features_check(skb, features);
2646 
2647 	/* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2648 	 * support inner IPv6 checksums and segmentation so  we need to
2649 	 * strip that feature if this is an IPv6 encapsulated frame.
2650 	 */
2651 	if (skb->encapsulation &&
2652 	    (skb->ip_summed == CHECKSUM_PARTIAL)) {
2653 		struct mlx4_en_priv *priv = netdev_priv(dev);
2654 
2655 		if (!priv->vxlan_port ||
2656 		    (ip_hdr(skb)->version != 4) ||
2657 		    (udp_hdr(skb)->dest != priv->vxlan_port))
2658 			features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2659 	}
2660 
2661 	return features;
2662 }
2663 
2664 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2665 {
2666 	struct mlx4_en_priv *priv = netdev_priv(dev);
2667 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2668 	struct mlx4_update_qp_params params;
2669 	int err;
2670 
2671 	if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2672 		return -EOPNOTSUPP;
2673 
2674 	/* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2675 	if (maxrate >> 12) {
2676 		params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2677 		params.rate_val  = maxrate / 1000;
2678 	} else if (maxrate) {
2679 		params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2680 		params.rate_val  = maxrate;
2681 	} else { /* zero serves to revoke the QP rate-limitation */
2682 		params.rate_unit = 0;
2683 		params.rate_val  = 0;
2684 	}
2685 
2686 	err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2687 			     &params);
2688 	return err;
2689 }
2690 
2691 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2692 {
2693 	struct mlx4_en_priv *priv = netdev_priv(dev);
2694 	struct mlx4_en_dev *mdev = priv->mdev;
2695 	struct mlx4_en_port_profile new_prof;
2696 	struct bpf_prog *old_prog;
2697 	struct mlx4_en_priv *tmp;
2698 	int tx_changed = 0;
2699 	int xdp_ring_num;
2700 	int port_up = 0;
2701 	int err;
2702 	int i;
2703 
2704 	xdp_ring_num = prog ? priv->rx_ring_num : 0;
2705 
2706 	/* No need to reconfigure buffers when simply swapping the
2707 	 * program for a new one.
2708 	 */
2709 	if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2710 		if (prog) {
2711 			prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2712 			if (IS_ERR(prog))
2713 				return PTR_ERR(prog);
2714 		}
2715 		mutex_lock(&mdev->state_lock);
2716 		for (i = 0; i < priv->rx_ring_num; i++) {
2717 			old_prog = rcu_dereference_protected(
2718 					priv->rx_ring[i]->xdp_prog,
2719 					lockdep_is_held(&mdev->state_lock));
2720 			rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2721 			if (old_prog)
2722 				bpf_prog_put(old_prog);
2723 		}
2724 		mutex_unlock(&mdev->state_lock);
2725 		return 0;
2726 	}
2727 
2728 	if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2729 		return -EOPNOTSUPP;
2730 
2731 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2732 	if (!tmp)
2733 		return -ENOMEM;
2734 
2735 	if (prog) {
2736 		prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2737 		if (IS_ERR(prog)) {
2738 			err = PTR_ERR(prog);
2739 			goto out;
2740 		}
2741 	}
2742 
2743 	mutex_lock(&mdev->state_lock);
2744 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2745 	new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2746 
2747 	if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2748 		tx_changed = 1;
2749 		new_prof.tx_ring_num[TX] =
2750 			MAX_TX_RINGS - ALIGN(xdp_ring_num, MLX4_EN_NUM_UP);
2751 		en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2752 	}
2753 
2754 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof);
2755 	if (err) {
2756 		if (prog)
2757 			bpf_prog_sub(prog, priv->rx_ring_num - 1);
2758 		goto unlock_out;
2759 	}
2760 
2761 	if (priv->port_up) {
2762 		port_up = 1;
2763 		mlx4_en_stop_port(dev, 1);
2764 	}
2765 
2766 	mlx4_en_safe_replace_resources(priv, tmp);
2767 	if (tx_changed)
2768 		netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2769 
2770 	for (i = 0; i < priv->rx_ring_num; i++) {
2771 		old_prog = rcu_dereference_protected(
2772 					priv->rx_ring[i]->xdp_prog,
2773 					lockdep_is_held(&mdev->state_lock));
2774 		rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2775 		if (old_prog)
2776 			bpf_prog_put(old_prog);
2777 	}
2778 
2779 	if (port_up) {
2780 		err = mlx4_en_start_port(dev);
2781 		if (err) {
2782 			en_err(priv, "Failed starting port %d for XDP change\n",
2783 			       priv->port);
2784 			queue_work(mdev->workqueue, &priv->watchdog_task);
2785 		}
2786 	}
2787 
2788 unlock_out:
2789 	mutex_unlock(&mdev->state_lock);
2790 out:
2791 	kfree(tmp);
2792 	return err;
2793 }
2794 
2795 static bool mlx4_xdp_attached(struct net_device *dev)
2796 {
2797 	struct mlx4_en_priv *priv = netdev_priv(dev);
2798 
2799 	return !!priv->tx_ring_num[TX_XDP];
2800 }
2801 
2802 static int mlx4_xdp(struct net_device *dev, struct netdev_xdp *xdp)
2803 {
2804 	switch (xdp->command) {
2805 	case XDP_SETUP_PROG:
2806 		return mlx4_xdp_set(dev, xdp->prog);
2807 	case XDP_QUERY_PROG:
2808 		xdp->prog_attached = mlx4_xdp_attached(dev);
2809 		return 0;
2810 	default:
2811 		return -EINVAL;
2812 	}
2813 }
2814 
2815 static const struct net_device_ops mlx4_netdev_ops = {
2816 	.ndo_open		= mlx4_en_open,
2817 	.ndo_stop		= mlx4_en_close,
2818 	.ndo_start_xmit		= mlx4_en_xmit,
2819 	.ndo_select_queue	= mlx4_en_select_queue,
2820 	.ndo_get_stats64	= mlx4_en_get_stats64,
2821 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2822 	.ndo_set_mac_address	= mlx4_en_set_mac,
2823 	.ndo_validate_addr	= eth_validate_addr,
2824 	.ndo_change_mtu		= mlx4_en_change_mtu,
2825 	.ndo_do_ioctl		= mlx4_en_ioctl,
2826 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2827 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2828 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2829 #ifdef CONFIG_NET_POLL_CONTROLLER
2830 	.ndo_poll_controller	= mlx4_en_netpoll,
2831 #endif
2832 	.ndo_set_features	= mlx4_en_set_features,
2833 	.ndo_fix_features	= mlx4_en_fix_features,
2834 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2835 #ifdef CONFIG_RFS_ACCEL
2836 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2837 #endif
2838 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2839 	.ndo_udp_tunnel_add	= mlx4_en_add_vxlan_port,
2840 	.ndo_udp_tunnel_del	= mlx4_en_del_vxlan_port,
2841 	.ndo_features_check	= mlx4_en_features_check,
2842 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2843 	.ndo_xdp		= mlx4_xdp,
2844 };
2845 
2846 static const struct net_device_ops mlx4_netdev_ops_master = {
2847 	.ndo_open		= mlx4_en_open,
2848 	.ndo_stop		= mlx4_en_close,
2849 	.ndo_start_xmit		= mlx4_en_xmit,
2850 	.ndo_select_queue	= mlx4_en_select_queue,
2851 	.ndo_get_stats64	= mlx4_en_get_stats64,
2852 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2853 	.ndo_set_mac_address	= mlx4_en_set_mac,
2854 	.ndo_validate_addr	= eth_validate_addr,
2855 	.ndo_change_mtu		= mlx4_en_change_mtu,
2856 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2857 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2858 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2859 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2860 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2861 	.ndo_set_vf_rate	= mlx4_en_set_vf_rate,
2862 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2863 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2864 	.ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2865 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2866 #ifdef CONFIG_NET_POLL_CONTROLLER
2867 	.ndo_poll_controller	= mlx4_en_netpoll,
2868 #endif
2869 	.ndo_set_features	= mlx4_en_set_features,
2870 	.ndo_fix_features	= mlx4_en_fix_features,
2871 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2872 #ifdef CONFIG_RFS_ACCEL
2873 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2874 #endif
2875 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2876 	.ndo_udp_tunnel_add	= mlx4_en_add_vxlan_port,
2877 	.ndo_udp_tunnel_del	= mlx4_en_del_vxlan_port,
2878 	.ndo_features_check	= mlx4_en_features_check,
2879 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2880 	.ndo_xdp		= mlx4_xdp,
2881 };
2882 
2883 struct mlx4_en_bond {
2884 	struct work_struct work;
2885 	struct mlx4_en_priv *priv;
2886 	int is_bonded;
2887 	struct mlx4_port_map port_map;
2888 };
2889 
2890 static void mlx4_en_bond_work(struct work_struct *work)
2891 {
2892 	struct mlx4_en_bond *bond = container_of(work,
2893 						     struct mlx4_en_bond,
2894 						     work);
2895 	int err = 0;
2896 	struct mlx4_dev *dev = bond->priv->mdev->dev;
2897 
2898 	if (bond->is_bonded) {
2899 		if (!mlx4_is_bonded(dev)) {
2900 			err = mlx4_bond(dev);
2901 			if (err)
2902 				en_err(bond->priv, "Fail to bond device\n");
2903 		}
2904 		if (!err) {
2905 			err = mlx4_port_map_set(dev, &bond->port_map);
2906 			if (err)
2907 				en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2908 				       bond->port_map.port1,
2909 				       bond->port_map.port2,
2910 				       err);
2911 		}
2912 	} else if (mlx4_is_bonded(dev)) {
2913 		err = mlx4_unbond(dev);
2914 		if (err)
2915 			en_err(bond->priv, "Fail to unbond device\n");
2916 	}
2917 	dev_put(bond->priv->dev);
2918 	kfree(bond);
2919 }
2920 
2921 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2922 				   u8 v2p_p1, u8 v2p_p2)
2923 {
2924 	struct mlx4_en_bond *bond = NULL;
2925 
2926 	bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2927 	if (!bond)
2928 		return -ENOMEM;
2929 
2930 	INIT_WORK(&bond->work, mlx4_en_bond_work);
2931 	bond->priv = priv;
2932 	bond->is_bonded = is_bonded;
2933 	bond->port_map.port1 = v2p_p1;
2934 	bond->port_map.port2 = v2p_p2;
2935 	dev_hold(priv->dev);
2936 	queue_work(priv->mdev->workqueue, &bond->work);
2937 	return 0;
2938 }
2939 
2940 int mlx4_en_netdev_event(struct notifier_block *this,
2941 			 unsigned long event, void *ptr)
2942 {
2943 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2944 	u8 port = 0;
2945 	struct mlx4_en_dev *mdev;
2946 	struct mlx4_dev *dev;
2947 	int i, num_eth_ports = 0;
2948 	bool do_bond = true;
2949 	struct mlx4_en_priv *priv;
2950 	u8 v2p_port1 = 0;
2951 	u8 v2p_port2 = 0;
2952 
2953 	if (!net_eq(dev_net(ndev), &init_net))
2954 		return NOTIFY_DONE;
2955 
2956 	mdev = container_of(this, struct mlx4_en_dev, nb);
2957 	dev = mdev->dev;
2958 
2959 	/* Go into this mode only when two network devices set on two ports
2960 	 * of the same mlx4 device are slaves of the same bonding master
2961 	 */
2962 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2963 		++num_eth_ports;
2964 		if (!port && (mdev->pndev[i] == ndev))
2965 			port = i;
2966 		mdev->upper[i] = mdev->pndev[i] ?
2967 			netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2968 		/* condition not met: network device is a slave */
2969 		if (!mdev->upper[i])
2970 			do_bond = false;
2971 		if (num_eth_ports < 2)
2972 			continue;
2973 		/* condition not met: same master */
2974 		if (mdev->upper[i] != mdev->upper[i-1])
2975 			do_bond = false;
2976 	}
2977 	/* condition not met: 2 salves */
2978 	do_bond = (num_eth_ports ==  2) ? do_bond : false;
2979 
2980 	/* handle only events that come with enough info */
2981 	if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
2982 		return NOTIFY_DONE;
2983 
2984 	priv = netdev_priv(ndev);
2985 	if (do_bond) {
2986 		struct netdev_notifier_bonding_info *notifier_info = ptr;
2987 		struct netdev_bonding_info *bonding_info =
2988 			&notifier_info->bonding_info;
2989 
2990 		/* required mode 1, 2 or 4 */
2991 		if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
2992 		    (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
2993 		    (bonding_info->master.bond_mode != BOND_MODE_8023AD))
2994 			do_bond = false;
2995 
2996 		/* require exactly 2 slaves */
2997 		if (bonding_info->master.num_slaves != 2)
2998 			do_bond = false;
2999 
3000 		/* calc v2p */
3001 		if (do_bond) {
3002 			if (bonding_info->master.bond_mode ==
3003 			    BOND_MODE_ACTIVEBACKUP) {
3004 				/* in active-backup mode virtual ports are
3005 				 * mapped to the physical port of the active
3006 				 * slave */
3007 				if (bonding_info->slave.state ==
3008 				    BOND_STATE_BACKUP) {
3009 					if (port == 1) {
3010 						v2p_port1 = 2;
3011 						v2p_port2 = 2;
3012 					} else {
3013 						v2p_port1 = 1;
3014 						v2p_port2 = 1;
3015 					}
3016 				} else { /* BOND_STATE_ACTIVE */
3017 					if (port == 1) {
3018 						v2p_port1 = 1;
3019 						v2p_port2 = 1;
3020 					} else {
3021 						v2p_port1 = 2;
3022 						v2p_port2 = 2;
3023 					}
3024 				}
3025 			} else { /* Active-Active */
3026 				/* in active-active mode a virtual port is
3027 				 * mapped to the native physical port if and only
3028 				 * if the physical port is up */
3029 				__s8 link = bonding_info->slave.link;
3030 
3031 				if (port == 1)
3032 					v2p_port2 = 2;
3033 				else
3034 					v2p_port1 = 1;
3035 				if ((link == BOND_LINK_UP) ||
3036 				    (link == BOND_LINK_FAIL)) {
3037 					if (port == 1)
3038 						v2p_port1 = 1;
3039 					else
3040 						v2p_port2 = 2;
3041 				} else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3042 					if (port == 1)
3043 						v2p_port1 = 2;
3044 					else
3045 						v2p_port2 = 1;
3046 				}
3047 			}
3048 		}
3049 	}
3050 
3051 	mlx4_en_queue_bond_work(priv, do_bond,
3052 				v2p_port1, v2p_port2);
3053 
3054 	return NOTIFY_DONE;
3055 }
3056 
3057 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3058 				     struct mlx4_en_stats_bitmap *stats_bitmap,
3059 				     u8 rx_ppp, u8 rx_pause,
3060 				     u8 tx_ppp, u8 tx_pause)
3061 {
3062 	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3063 
3064 	if (!mlx4_is_slave(dev) &&
3065 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3066 		mutex_lock(&stats_bitmap->mutex);
3067 		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3068 
3069 		if (rx_ppp)
3070 			bitmap_set(stats_bitmap->bitmap, last_i,
3071 				   NUM_FLOW_PRIORITY_STATS_RX);
3072 		last_i += NUM_FLOW_PRIORITY_STATS_RX;
3073 
3074 		if (rx_pause && !(rx_ppp))
3075 			bitmap_set(stats_bitmap->bitmap, last_i,
3076 				   NUM_FLOW_STATS_RX);
3077 		last_i += NUM_FLOW_STATS_RX;
3078 
3079 		if (tx_ppp)
3080 			bitmap_set(stats_bitmap->bitmap, last_i,
3081 				   NUM_FLOW_PRIORITY_STATS_TX);
3082 		last_i += NUM_FLOW_PRIORITY_STATS_TX;
3083 
3084 		if (tx_pause && !(tx_ppp))
3085 			bitmap_set(stats_bitmap->bitmap, last_i,
3086 				   NUM_FLOW_STATS_TX);
3087 		last_i += NUM_FLOW_STATS_TX;
3088 
3089 		mutex_unlock(&stats_bitmap->mutex);
3090 	}
3091 }
3092 
3093 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3094 			      struct mlx4_en_stats_bitmap *stats_bitmap,
3095 			      u8 rx_ppp, u8 rx_pause,
3096 			      u8 tx_ppp, u8 tx_pause)
3097 {
3098 	int last_i = 0;
3099 
3100 	mutex_init(&stats_bitmap->mutex);
3101 	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3102 
3103 	if (mlx4_is_slave(dev)) {
3104 		bitmap_set(stats_bitmap->bitmap, last_i +
3105 					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3106 		bitmap_set(stats_bitmap->bitmap, last_i +
3107 					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3108 		bitmap_set(stats_bitmap->bitmap, last_i +
3109 					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3110 		bitmap_set(stats_bitmap->bitmap, last_i +
3111 					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3112 		bitmap_set(stats_bitmap->bitmap, last_i +
3113 					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3114 		bitmap_set(stats_bitmap->bitmap, last_i +
3115 					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3116 	} else {
3117 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3118 	}
3119 	last_i += NUM_MAIN_STATS;
3120 
3121 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3122 	last_i += NUM_PORT_STATS;
3123 
3124 	if (mlx4_is_master(dev))
3125 		bitmap_set(stats_bitmap->bitmap, last_i,
3126 			   NUM_PF_STATS);
3127 	last_i += NUM_PF_STATS;
3128 
3129 	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3130 					rx_ppp, rx_pause,
3131 					tx_ppp, tx_pause);
3132 	last_i += NUM_FLOW_STATS;
3133 
3134 	if (!mlx4_is_slave(dev))
3135 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3136 	last_i += NUM_PKT_STATS;
3137 
3138 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3139 	last_i += NUM_XDP_STATS;
3140 }
3141 
3142 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3143 			struct mlx4_en_port_profile *prof)
3144 {
3145 	struct net_device *dev;
3146 	struct mlx4_en_priv *priv;
3147 	int i, t;
3148 	int err;
3149 
3150 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3151 				 MAX_TX_RINGS, MAX_RX_RINGS);
3152 	if (dev == NULL)
3153 		return -ENOMEM;
3154 
3155 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3156 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3157 
3158 	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3159 	dev->dev_port = port - 1;
3160 
3161 	/*
3162 	 * Initialize driver private data
3163 	 */
3164 
3165 	priv = netdev_priv(dev);
3166 	memset(priv, 0, sizeof(struct mlx4_en_priv));
3167 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3168 	spin_lock_init(&priv->stats_lock);
3169 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3170 	INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
3171 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3172 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3173 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3174 	INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
3175 	INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
3176 #ifdef CONFIG_RFS_ACCEL
3177 	INIT_LIST_HEAD(&priv->filters);
3178 	spin_lock_init(&priv->filters_lock);
3179 #endif
3180 
3181 	priv->dev = dev;
3182 	priv->mdev = mdev;
3183 	priv->ddev = &mdev->pdev->dev;
3184 	priv->prof = prof;
3185 	priv->port = port;
3186 	priv->port_up = false;
3187 	priv->flags = prof->flags;
3188 	priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3189 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3190 			MLX4_WQE_CTRL_SOLICITED);
3191 	priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
3192 	priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3193 	netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3194 
3195 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3196 		priv->tx_ring_num[t] = prof->tx_ring_num[t];
3197 		if (!priv->tx_ring_num[t])
3198 			continue;
3199 
3200 		priv->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
3201 					   MAX_TX_RINGS, GFP_KERNEL);
3202 		if (!priv->tx_ring[t]) {
3203 			err = -ENOMEM;
3204 			goto err_free_tx;
3205 		}
3206 		priv->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
3207 					 MAX_TX_RINGS, GFP_KERNEL);
3208 		if (!priv->tx_cq[t]) {
3209 			kfree(priv->tx_ring[t]);
3210 			err = -ENOMEM;
3211 			goto out;
3212 		}
3213 	}
3214 	priv->rx_ring_num = prof->rx_ring_num;
3215 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3216 	priv->cqe_size = mdev->dev->caps.cqe_size;
3217 	priv->mac_index = -1;
3218 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
3219 #ifdef CONFIG_MLX4_EN_DCB
3220 	if (!mlx4_is_slave(priv->mdev->dev)) {
3221 		priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3222 			DCB_CAP_DCBX_VER_IEEE;
3223 		priv->flags |= MLX4_EN_DCB_ENABLED;
3224 		priv->cee_config.pfc_state = false;
3225 
3226 		for (i = 0; i < MLX4_EN_NUM_UP; i++)
3227 			priv->cee_config.dcb_pfc[i] = pfc_disabled;
3228 
3229 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3230 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3231 		} else {
3232 			en_info(priv, "enabling only PFC DCB ops\n");
3233 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3234 		}
3235 	}
3236 #endif
3237 
3238 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3239 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
3240 
3241 	/* Query for default mac and max mtu */
3242 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3243 
3244 	if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3245 	    MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3246 		priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3247 
3248 	/* Set default MAC */
3249 	dev->addr_len = ETH_ALEN;
3250 	mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3251 	if (!is_valid_ether_addr(dev->dev_addr)) {
3252 		en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
3253 		       priv->port, dev->dev_addr);
3254 		err = -EINVAL;
3255 		goto out;
3256 	} else if (mlx4_is_slave(priv->mdev->dev) &&
3257 		   (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3258 		/* Random MAC was assigned in mlx4_slave_cap
3259 		 * in mlx4_core module
3260 		 */
3261 		dev->addr_assign_type |= NET_ADDR_RANDOM;
3262 		en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3263 	}
3264 
3265 	memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3266 
3267 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3268 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3269 	err = mlx4_en_alloc_resources(priv);
3270 	if (err)
3271 		goto out;
3272 
3273 	/* Initialize time stamping config */
3274 	priv->hwtstamp_config.flags = 0;
3275 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3276 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3277 
3278 	/* Allocate page for receive rings */
3279 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3280 				MLX4_EN_PAGE_SIZE);
3281 	if (err) {
3282 		en_err(priv, "Failed to allocate page for rx qps\n");
3283 		goto out;
3284 	}
3285 	priv->allocated = 1;
3286 
3287 	/*
3288 	 * Initialize netdev entry points
3289 	 */
3290 	if (mlx4_is_master(priv->mdev->dev))
3291 		dev->netdev_ops = &mlx4_netdev_ops_master;
3292 	else
3293 		dev->netdev_ops = &mlx4_netdev_ops;
3294 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3295 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3296 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3297 
3298 	dev->ethtool_ops = &mlx4_en_ethtool_ops;
3299 
3300 	/*
3301 	 * Set driver features
3302 	 */
3303 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3304 	if (mdev->LSO_support)
3305 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3306 
3307 	dev->vlan_features = dev->hw_features;
3308 
3309 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3310 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3311 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3312 			NETIF_F_HW_VLAN_CTAG_FILTER;
3313 	dev->hw_features |= NETIF_F_LOOPBACK |
3314 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3315 
3316 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3317 		dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3318 			NETIF_F_HW_VLAN_STAG_FILTER;
3319 		dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3320 	}
3321 
3322 	if (mlx4_is_slave(mdev->dev)) {
3323 		bool vlan_offload_disabled;
3324 		int phv;
3325 
3326 		err = get_phv_bit(mdev->dev, port, &phv);
3327 		if (!err && phv) {
3328 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3329 			priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3330 		}
3331 		err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3332 							&vlan_offload_disabled);
3333 		if (!err && vlan_offload_disabled) {
3334 			dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3335 					      NETIF_F_HW_VLAN_CTAG_RX |
3336 					      NETIF_F_HW_VLAN_STAG_TX |
3337 					      NETIF_F_HW_VLAN_STAG_RX);
3338 			dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3339 					   NETIF_F_HW_VLAN_CTAG_RX |
3340 					   NETIF_F_HW_VLAN_STAG_TX |
3341 					   NETIF_F_HW_VLAN_STAG_RX);
3342 		}
3343 	} else {
3344 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3345 		    !(mdev->dev->caps.flags2 &
3346 		      MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3347 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3348 	}
3349 
3350 	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3351 		dev->hw_features |= NETIF_F_RXFCS;
3352 
3353 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3354 		dev->hw_features |= NETIF_F_RXALL;
3355 
3356 	if (mdev->dev->caps.steering_mode ==
3357 	    MLX4_STEERING_MODE_DEVICE_MANAGED &&
3358 	    mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3359 		dev->hw_features |= NETIF_F_NTUPLE;
3360 
3361 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3362 		dev->priv_flags |= IFF_UNICAST_FLT;
3363 
3364 	/* Setting a default hash function value */
3365 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3366 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3367 	} else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3368 		priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3369 	} else {
3370 		en_warn(priv,
3371 			"No RSS hash capabilities exposed, using Toeplitz\n");
3372 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3373 	}
3374 
3375 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3376 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3377 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3378 				    NETIF_F_GSO_PARTIAL;
3379 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3380 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3381 				    NETIF_F_GSO_PARTIAL;
3382 		dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3383 	}
3384 
3385 	/* MTU range: 46 - hw-specific max */
3386 	dev->min_mtu = MLX4_EN_MIN_MTU;
3387 	dev->max_mtu = priv->max_mtu;
3388 
3389 	mdev->pndev[port] = dev;
3390 	mdev->upper[port] = NULL;
3391 
3392 	netif_carrier_off(dev);
3393 	mlx4_en_set_default_moderation(priv);
3394 
3395 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3396 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3397 
3398 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3399 
3400 	/* Configure port */
3401 	mlx4_en_calc_rx_buf(dev);
3402 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3403 				    priv->rx_skb_size + ETH_FCS_LEN,
3404 				    prof->tx_pause, prof->tx_ppp,
3405 				    prof->rx_pause, prof->rx_ppp);
3406 	if (err) {
3407 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3408 		       priv->port, err);
3409 		goto out;
3410 	}
3411 
3412 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3413 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3414 		if (err) {
3415 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3416 			       err);
3417 			goto out;
3418 		}
3419 	}
3420 
3421 	/* Init port */
3422 	en_warn(priv, "Initializing port\n");
3423 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
3424 	if (err) {
3425 		en_err(priv, "Failed Initializing port\n");
3426 		goto out;
3427 	}
3428 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3429 
3430 	/* Initialize time stamp mechanism */
3431 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3432 		mlx4_en_init_timestamp(mdev);
3433 
3434 	queue_delayed_work(mdev->workqueue, &priv->service_task,
3435 			   SERVICE_TASK_DELAY);
3436 
3437 	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3438 				 mdev->profile.prof[priv->port].rx_ppp,
3439 				 mdev->profile.prof[priv->port].rx_pause,
3440 				 mdev->profile.prof[priv->port].tx_ppp,
3441 				 mdev->profile.prof[priv->port].tx_pause);
3442 
3443 	err = register_netdev(dev);
3444 	if (err) {
3445 		en_err(priv, "Netdev registration failed for port %d\n", port);
3446 		goto out;
3447 	}
3448 
3449 	priv->registered = 1;
3450 	devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3451 				  dev);
3452 
3453 	return 0;
3454 
3455 err_free_tx:
3456 	while (t--) {
3457 		kfree(priv->tx_ring[t]);
3458 		kfree(priv->tx_cq[t]);
3459 	}
3460 out:
3461 	mlx4_en_destroy_netdev(dev);
3462 	return err;
3463 }
3464 
3465 int mlx4_en_reset_config(struct net_device *dev,
3466 			 struct hwtstamp_config ts_config,
3467 			 netdev_features_t features)
3468 {
3469 	struct mlx4_en_priv *priv = netdev_priv(dev);
3470 	struct mlx4_en_dev *mdev = priv->mdev;
3471 	struct mlx4_en_port_profile new_prof;
3472 	struct mlx4_en_priv *tmp;
3473 	int port_up = 0;
3474 	int err = 0;
3475 
3476 	if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3477 	    priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3478 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3479 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3480 		return 0; /* Nothing to change */
3481 
3482 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3483 	    (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3484 	    (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3485 		en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3486 		return -EINVAL;
3487 	}
3488 
3489 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3490 	if (!tmp)
3491 		return -ENOMEM;
3492 
3493 	mutex_lock(&mdev->state_lock);
3494 
3495 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3496 	memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3497 
3498 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof);
3499 	if (err)
3500 		goto out;
3501 
3502 	if (priv->port_up) {
3503 		port_up = 1;
3504 		mlx4_en_stop_port(dev, 1);
3505 	}
3506 
3507 	en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
3508 		ts_config.rx_filter,
3509 		!!(features & NETIF_F_HW_VLAN_CTAG_RX));
3510 
3511 	mlx4_en_safe_replace_resources(priv, tmp);
3512 
3513 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3514 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3515 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3516 		else
3517 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3518 	} else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3519 		/* RX time-stamping is OFF, update the RX vlan offload
3520 		 * to the latest wanted state
3521 		 */
3522 		if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3523 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3524 		else
3525 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3526 	}
3527 
3528 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3529 		if (features & NETIF_F_RXFCS)
3530 			dev->features |= NETIF_F_RXFCS;
3531 		else
3532 			dev->features &= ~NETIF_F_RXFCS;
3533 	}
3534 
3535 	/* RX vlan offload and RX time-stamping can't co-exist !
3536 	 * Regardless of the caller's choice,
3537 	 * Turn Off RX vlan offload in case of time-stamping is ON
3538 	 */
3539 	if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3540 		if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3541 			en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3542 		dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3543 	}
3544 
3545 	if (port_up) {
3546 		err = mlx4_en_start_port(dev);
3547 		if (err)
3548 			en_err(priv, "Failed starting port\n");
3549 	}
3550 
3551 out:
3552 	mutex_unlock(&mdev->state_lock);
3553 	kfree(tmp);
3554 	if (!err)
3555 		netdev_features_change(dev);
3556 	return err;
3557 }
3558