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/etherdevice.h>
35 #include <linux/tcp.h>
36 #include <linux/if_vlan.h>
37 #include <linux/delay.h>
38 #include <linux/slab.h>
39 #include <linux/hash.h>
40 #include <net/ip.h>
41 
42 #include <linux/mlx4/driver.h>
43 #include <linux/mlx4/device.h>
44 #include <linux/mlx4/cmd.h>
45 #include <linux/mlx4/cq.h>
46 
47 #include "mlx4_en.h"
48 #include "en_port.h"
49 
50 static int mlx4_en_setup_tc(struct net_device *dev, u8 up)
51 {
52 	struct mlx4_en_priv *priv = netdev_priv(dev);
53 	int i;
54 	unsigned int q, offset = 0;
55 
56 	if (up && up != MLX4_EN_NUM_UP)
57 		return -EINVAL;
58 
59 	netdev_set_num_tc(dev, up);
60 
61 	/* Partition Tx queues evenly amongst UP's */
62 	q = priv->tx_ring_num / up;
63 	for (i = 0; i < up; i++) {
64 		netdev_set_tc_queue(dev, i, q, offset);
65 		offset += q;
66 	}
67 
68 	return 0;
69 }
70 
71 #ifdef CONFIG_RFS_ACCEL
72 
73 struct mlx4_en_filter {
74 	struct list_head next;
75 	struct work_struct work;
76 
77 	__be32 src_ip;
78 	__be32 dst_ip;
79 	__be16 src_port;
80 	__be16 dst_port;
81 
82 	int rxq_index;
83 	struct mlx4_en_priv *priv;
84 	u32 flow_id;			/* RFS infrastructure id */
85 	int id;				/* mlx4_en driver id */
86 	u64 reg_id;			/* Flow steering API id */
87 	u8 activated;			/* Used to prevent expiry before filter
88 					 * is attached
89 					 */
90 	struct hlist_node filter_chain;
91 };
92 
93 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
94 
95 static void mlx4_en_filter_work(struct work_struct *work)
96 {
97 	struct mlx4_en_filter *filter = container_of(work,
98 						     struct mlx4_en_filter,
99 						     work);
100 	struct mlx4_en_priv *priv = filter->priv;
101 	struct mlx4_spec_list spec_tcp = {
102 		.id = MLX4_NET_TRANS_RULE_ID_TCP,
103 		{
104 			.tcp_udp = {
105 				.dst_port = filter->dst_port,
106 				.dst_port_msk = (__force __be16)-1,
107 				.src_port = filter->src_port,
108 				.src_port_msk = (__force __be16)-1,
109 			},
110 		},
111 	};
112 	struct mlx4_spec_list spec_ip = {
113 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
114 		{
115 			.ipv4 = {
116 				.dst_ip = filter->dst_ip,
117 				.dst_ip_msk = (__force __be32)-1,
118 				.src_ip = filter->src_ip,
119 				.src_ip_msk = (__force __be32)-1,
120 			},
121 		},
122 	};
123 	struct mlx4_spec_list spec_eth = {
124 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
125 	};
126 	struct mlx4_net_trans_rule rule = {
127 		.list = LIST_HEAD_INIT(rule.list),
128 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
129 		.exclusive = 1,
130 		.allow_loopback = 1,
131 		.promisc_mode = MLX4_FS_PROMISC_NONE,
132 		.port = priv->port,
133 		.priority = MLX4_DOMAIN_RFS,
134 	};
135 	int rc;
136 	__be64 mac;
137 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
138 
139 	list_add_tail(&spec_eth.list, &rule.list);
140 	list_add_tail(&spec_ip.list, &rule.list);
141 	list_add_tail(&spec_tcp.list, &rule.list);
142 
143 	mac = cpu_to_be64((priv->mac & MLX4_MAC_MASK) << 16);
144 
145 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
146 	memcpy(spec_eth.eth.dst_mac, &mac, ETH_ALEN);
147 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
148 
149 	filter->activated = 0;
150 
151 	if (filter->reg_id) {
152 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
153 		if (rc && rc != -ENOENT)
154 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
155 	}
156 
157 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
158 	if (rc)
159 		en_err(priv, "Error attaching flow. err = %d\n", rc);
160 
161 	mlx4_en_filter_rfs_expire(priv);
162 
163 	filter->activated = 1;
164 }
165 
166 static inline struct hlist_head *
167 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
168 		   __be16 src_port, __be16 dst_port)
169 {
170 	unsigned long l;
171 	int bucket_idx;
172 
173 	l = (__force unsigned long)src_port |
174 	    ((__force unsigned long)dst_port << 2);
175 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
176 
177 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
178 
179 	return &priv->filter_hash[bucket_idx];
180 }
181 
182 static struct mlx4_en_filter *
183 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
184 		     __be32 dst_ip, __be16 src_port, __be16 dst_port,
185 		     u32 flow_id)
186 {
187 	struct mlx4_en_filter *filter = NULL;
188 
189 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
190 	if (!filter)
191 		return NULL;
192 
193 	filter->priv = priv;
194 	filter->rxq_index = rxq_index;
195 	INIT_WORK(&filter->work, mlx4_en_filter_work);
196 
197 	filter->src_ip = src_ip;
198 	filter->dst_ip = dst_ip;
199 	filter->src_port = src_port;
200 	filter->dst_port = dst_port;
201 
202 	filter->flow_id = flow_id;
203 
204 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
205 
206 	list_add_tail(&filter->next, &priv->filters);
207 	hlist_add_head(&filter->filter_chain,
208 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
209 					  dst_port));
210 
211 	return filter;
212 }
213 
214 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
215 {
216 	struct mlx4_en_priv *priv = filter->priv;
217 	int rc;
218 
219 	list_del(&filter->next);
220 
221 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
222 	if (rc && rc != -ENOENT)
223 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
224 
225 	kfree(filter);
226 }
227 
228 static inline struct mlx4_en_filter *
229 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
230 		    __be16 src_port, __be16 dst_port)
231 {
232 	struct hlist_node *elem;
233 	struct mlx4_en_filter *filter;
234 	struct mlx4_en_filter *ret = NULL;
235 
236 	hlist_for_each_entry(filter, elem,
237 			     filter_hash_bucket(priv, src_ip, dst_ip,
238 						src_port, dst_port),
239 			     filter_chain) {
240 		if (filter->src_ip == src_ip &&
241 		    filter->dst_ip == dst_ip &&
242 		    filter->src_port == src_port &&
243 		    filter->dst_port == dst_port) {
244 			ret = filter;
245 			break;
246 		}
247 	}
248 
249 	return ret;
250 }
251 
252 static int
253 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
254 		   u16 rxq_index, u32 flow_id)
255 {
256 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
257 	struct mlx4_en_filter *filter;
258 	const struct iphdr *ip;
259 	const __be16 *ports;
260 	__be32 src_ip;
261 	__be32 dst_ip;
262 	__be16 src_port;
263 	__be16 dst_port;
264 	int nhoff = skb_network_offset(skb);
265 	int ret = 0;
266 
267 	if (skb->protocol != htons(ETH_P_IP))
268 		return -EPROTONOSUPPORT;
269 
270 	ip = (const struct iphdr *)(skb->data + nhoff);
271 	if (ip_is_fragment(ip))
272 		return -EPROTONOSUPPORT;
273 
274 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
275 
276 	src_ip = ip->saddr;
277 	dst_ip = ip->daddr;
278 	src_port = ports[0];
279 	dst_port = ports[1];
280 
281 	if (ip->protocol != IPPROTO_TCP)
282 		return -EPROTONOSUPPORT;
283 
284 	spin_lock_bh(&priv->filters_lock);
285 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, src_port, dst_port);
286 	if (filter) {
287 		if (filter->rxq_index == rxq_index)
288 			goto out;
289 
290 		filter->rxq_index = rxq_index;
291 	} else {
292 		filter = mlx4_en_filter_alloc(priv, rxq_index,
293 					      src_ip, dst_ip,
294 					      src_port, dst_port, flow_id);
295 		if (!filter) {
296 			ret = -ENOMEM;
297 			goto err;
298 		}
299 	}
300 
301 	queue_work(priv->mdev->workqueue, &filter->work);
302 
303 out:
304 	ret = filter->id;
305 err:
306 	spin_unlock_bh(&priv->filters_lock);
307 
308 	return ret;
309 }
310 
311 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv,
312 			     struct mlx4_en_rx_ring *rx_ring)
313 {
314 	struct mlx4_en_filter *filter, *tmp;
315 	LIST_HEAD(del_list);
316 
317 	spin_lock_bh(&priv->filters_lock);
318 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
319 		list_move(&filter->next, &del_list);
320 		hlist_del(&filter->filter_chain);
321 	}
322 	spin_unlock_bh(&priv->filters_lock);
323 
324 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
325 		cancel_work_sync(&filter->work);
326 		mlx4_en_filter_free(filter);
327 	}
328 }
329 
330 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
331 {
332 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
333 	LIST_HEAD(del_list);
334 	int i = 0;
335 
336 	spin_lock_bh(&priv->filters_lock);
337 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
338 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
339 			break;
340 
341 		if (filter->activated &&
342 		    !work_pending(&filter->work) &&
343 		    rps_may_expire_flow(priv->dev,
344 					filter->rxq_index, filter->flow_id,
345 					filter->id)) {
346 			list_move(&filter->next, &del_list);
347 			hlist_del(&filter->filter_chain);
348 		} else
349 			last_filter = filter;
350 
351 		i++;
352 	}
353 
354 	if (last_filter && (&last_filter->next != priv->filters.next))
355 		list_move(&priv->filters, &last_filter->next);
356 
357 	spin_unlock_bh(&priv->filters_lock);
358 
359 	list_for_each_entry_safe(filter, tmp, &del_list, next)
360 		mlx4_en_filter_free(filter);
361 }
362 #endif
363 
364 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev, unsigned short vid)
365 {
366 	struct mlx4_en_priv *priv = netdev_priv(dev);
367 	struct mlx4_en_dev *mdev = priv->mdev;
368 	int err;
369 	int idx;
370 
371 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
372 
373 	set_bit(vid, priv->active_vlans);
374 
375 	/* Add VID to port VLAN filter */
376 	mutex_lock(&mdev->state_lock);
377 	if (mdev->device_up && priv->port_up) {
378 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
379 		if (err)
380 			en_err(priv, "Failed configuring VLAN filter\n");
381 	}
382 	if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
383 		en_err(priv, "failed adding vlan %d\n", vid);
384 	mutex_unlock(&mdev->state_lock);
385 
386 	return 0;
387 }
388 
389 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
390 {
391 	struct mlx4_en_priv *priv = netdev_priv(dev);
392 	struct mlx4_en_dev *mdev = priv->mdev;
393 	int err;
394 	int idx;
395 
396 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
397 
398 	clear_bit(vid, priv->active_vlans);
399 
400 	/* Remove VID from port VLAN filter */
401 	mutex_lock(&mdev->state_lock);
402 	if (!mlx4_find_cached_vlan(mdev->dev, priv->port, vid, &idx))
403 		mlx4_unregister_vlan(mdev->dev, priv->port, idx);
404 	else
405 		en_err(priv, "could not find vid %d in cache\n", vid);
406 
407 	if (mdev->device_up && priv->port_up) {
408 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
409 		if (err)
410 			en_err(priv, "Failed configuring VLAN filter\n");
411 	}
412 	mutex_unlock(&mdev->state_lock);
413 
414 	return 0;
415 }
416 
417 u64 mlx4_en_mac_to_u64(u8 *addr)
418 {
419 	u64 mac = 0;
420 	int i;
421 
422 	for (i = 0; i < ETH_ALEN; i++) {
423 		mac <<= 8;
424 		mac |= addr[i];
425 	}
426 	return mac;
427 }
428 
429 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
430 {
431 	struct mlx4_en_priv *priv = netdev_priv(dev);
432 	struct mlx4_en_dev *mdev = priv->mdev;
433 	struct sockaddr *saddr = addr;
434 
435 	if (!is_valid_ether_addr(saddr->sa_data))
436 		return -EADDRNOTAVAIL;
437 
438 	memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
439 	priv->mac = mlx4_en_mac_to_u64(dev->dev_addr);
440 	queue_work(mdev->workqueue, &priv->mac_task);
441 	return 0;
442 }
443 
444 static void mlx4_en_do_set_mac(struct work_struct *work)
445 {
446 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
447 						 mac_task);
448 	struct mlx4_en_dev *mdev = priv->mdev;
449 	int err = 0;
450 
451 	mutex_lock(&mdev->state_lock);
452 	if (priv->port_up) {
453 		/* Remove old MAC and insert the new one */
454 		err = mlx4_replace_mac(mdev->dev, priv->port,
455 				       priv->base_qpn, priv->mac);
456 		if (err)
457 			en_err(priv, "Failed changing HW MAC address\n");
458 	} else
459 		en_dbg(HW, priv, "Port is down while "
460 				 "registering mac, exiting...\n");
461 
462 	mutex_unlock(&mdev->state_lock);
463 }
464 
465 static void mlx4_en_clear_list(struct net_device *dev)
466 {
467 	struct mlx4_en_priv *priv = netdev_priv(dev);
468 	struct mlx4_en_mc_list *tmp, *mc_to_del;
469 
470 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
471 		list_del(&mc_to_del->list);
472 		kfree(mc_to_del);
473 	}
474 }
475 
476 static void mlx4_en_cache_mclist(struct net_device *dev)
477 {
478 	struct mlx4_en_priv *priv = netdev_priv(dev);
479 	struct netdev_hw_addr *ha;
480 	struct mlx4_en_mc_list *tmp;
481 
482 	mlx4_en_clear_list(dev);
483 	netdev_for_each_mc_addr(ha, dev) {
484 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
485 		if (!tmp) {
486 			en_err(priv, "failed to allocate multicast list\n");
487 			mlx4_en_clear_list(dev);
488 			return;
489 		}
490 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
491 		list_add_tail(&tmp->list, &priv->mc_list);
492 	}
493 }
494 
495 static void update_mclist_flags(struct mlx4_en_priv *priv,
496 				struct list_head *dst,
497 				struct list_head *src)
498 {
499 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
500 	bool found;
501 
502 	/* Find all the entries that should be removed from dst,
503 	 * These are the entries that are not found in src
504 	 */
505 	list_for_each_entry(dst_tmp, dst, list) {
506 		found = false;
507 		list_for_each_entry(src_tmp, src, list) {
508 			if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
509 				found = true;
510 				break;
511 			}
512 		}
513 		if (!found)
514 			dst_tmp->action = MCLIST_REM;
515 	}
516 
517 	/* Add entries that exist in src but not in dst
518 	 * mark them as need to add
519 	 */
520 	list_for_each_entry(src_tmp, src, list) {
521 		found = false;
522 		list_for_each_entry(dst_tmp, dst, list) {
523 			if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
524 				dst_tmp->action = MCLIST_NONE;
525 				found = true;
526 				break;
527 			}
528 		}
529 		if (!found) {
530 			new_mc = kmalloc(sizeof(struct mlx4_en_mc_list),
531 					 GFP_KERNEL);
532 			if (!new_mc) {
533 				en_err(priv, "Failed to allocate current multicast list\n");
534 				return;
535 			}
536 			memcpy(new_mc, src_tmp,
537 			       sizeof(struct mlx4_en_mc_list));
538 			new_mc->action = MCLIST_ADD;
539 			list_add_tail(&new_mc->list, dst);
540 		}
541 	}
542 }
543 
544 static void mlx4_en_set_multicast(struct net_device *dev)
545 {
546 	struct mlx4_en_priv *priv = netdev_priv(dev);
547 
548 	if (!priv->port_up)
549 		return;
550 
551 	queue_work(priv->mdev->workqueue, &priv->mcast_task);
552 }
553 
554 static void mlx4_en_do_set_multicast(struct work_struct *work)
555 {
556 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
557 						 mcast_task);
558 	struct mlx4_en_dev *mdev = priv->mdev;
559 	struct net_device *dev = priv->dev;
560 	struct mlx4_en_mc_list *mclist, *tmp;
561 	u64 mcast_addr = 0;
562 	u8 mc_list[16] = {0};
563 	int err = 0;
564 
565 	mutex_lock(&mdev->state_lock);
566 	if (!mdev->device_up) {
567 		en_dbg(HW, priv, "Card is not up, "
568 				 "ignoring multicast change.\n");
569 		goto out;
570 	}
571 	if (!priv->port_up) {
572 		en_dbg(HW, priv, "Port is down, "
573 				 "ignoring  multicast change.\n");
574 		goto out;
575 	}
576 
577 	if (!netif_carrier_ok(dev)) {
578 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
579 			if (priv->port_state.link_state) {
580 				priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
581 				netif_carrier_on(dev);
582 				en_dbg(LINK, priv, "Link Up\n");
583 			}
584 		}
585 	}
586 
587 	/*
588 	 * Promsicuous mode: disable all filters
589 	 */
590 
591 	if (dev->flags & IFF_PROMISC) {
592 		if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
593 			if (netif_msg_rx_status(priv))
594 				en_warn(priv, "Entering promiscuous mode\n");
595 			priv->flags |= MLX4_EN_FLAG_PROMISC;
596 
597 			/* Enable promiscouos mode */
598 			switch (mdev->dev->caps.steering_mode) {
599 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
600 				err = mlx4_flow_steer_promisc_add(mdev->dev,
601 								  priv->port,
602 								  priv->base_qpn,
603 								  MLX4_FS_PROMISC_UPLINK);
604 				if (err)
605 					en_err(priv, "Failed enabling promiscuous mode\n");
606 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
607 				break;
608 
609 			case MLX4_STEERING_MODE_B0:
610 				err = mlx4_unicast_promisc_add(mdev->dev,
611 							       priv->base_qpn,
612 							       priv->port);
613 				if (err)
614 					en_err(priv, "Failed enabling unicast promiscuous mode\n");
615 
616 				/* Add the default qp number as multicast
617 				 * promisc
618 				 */
619 				if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
620 					err = mlx4_multicast_promisc_add(mdev->dev,
621 									 priv->base_qpn,
622 									 priv->port);
623 					if (err)
624 						en_err(priv, "Failed enabling multicast promiscuous mode\n");
625 					priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
626 				}
627 				break;
628 
629 			case MLX4_STEERING_MODE_A0:
630 				err = mlx4_SET_PORT_qpn_calc(mdev->dev,
631 							     priv->port,
632 							     priv->base_qpn,
633 							     1);
634 				if (err)
635 					en_err(priv, "Failed enabling promiscuous mode\n");
636 				break;
637 			}
638 
639 			/* Disable port multicast filter (unconditionally) */
640 			err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
641 						  0, MLX4_MCAST_DISABLE);
642 			if (err)
643 				en_err(priv, "Failed disabling "
644 					     "multicast filter\n");
645 
646 			/* Disable port VLAN filter */
647 			err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
648 			if (err)
649 				en_err(priv, "Failed disabling VLAN filter\n");
650 		}
651 		goto out;
652 	}
653 
654 	/*
655 	 * Not in promiscuous mode
656 	 */
657 
658 	if (priv->flags & MLX4_EN_FLAG_PROMISC) {
659 		if (netif_msg_rx_status(priv))
660 			en_warn(priv, "Leaving promiscuous mode\n");
661 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
662 
663 		/* Disable promiscouos mode */
664 		switch (mdev->dev->caps.steering_mode) {
665 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
666 			err = mlx4_flow_steer_promisc_remove(mdev->dev,
667 							     priv->port,
668 							     MLX4_FS_PROMISC_UPLINK);
669 			if (err)
670 				en_err(priv, "Failed disabling promiscuous mode\n");
671 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
672 			break;
673 
674 		case MLX4_STEERING_MODE_B0:
675 			err = mlx4_unicast_promisc_remove(mdev->dev,
676 							  priv->base_qpn,
677 							  priv->port);
678 			if (err)
679 				en_err(priv, "Failed disabling unicast promiscuous mode\n");
680 			/* Disable Multicast promisc */
681 			if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
682 				err = mlx4_multicast_promisc_remove(mdev->dev,
683 								    priv->base_qpn,
684 								    priv->port);
685 				if (err)
686 					en_err(priv, "Failed disabling multicast promiscuous mode\n");
687 				priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
688 			}
689 			break;
690 
691 		case MLX4_STEERING_MODE_A0:
692 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
693 						     priv->port,
694 						     priv->base_qpn, 0);
695 			if (err)
696 				en_err(priv, "Failed disabling promiscuous mode\n");
697 			break;
698 		}
699 
700 		/* Enable port VLAN filter */
701 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
702 		if (err)
703 			en_err(priv, "Failed enabling VLAN filter\n");
704 	}
705 
706 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
707 	if (dev->flags & IFF_ALLMULTI) {
708 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
709 					  0, MLX4_MCAST_DISABLE);
710 		if (err)
711 			en_err(priv, "Failed disabling multicast filter\n");
712 
713 		/* Add the default qp number as multicast promisc */
714 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
715 			switch (mdev->dev->caps.steering_mode) {
716 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
717 				err = mlx4_flow_steer_promisc_add(mdev->dev,
718 								  priv->port,
719 								  priv->base_qpn,
720 								  MLX4_FS_PROMISC_ALL_MULTI);
721 				break;
722 
723 			case MLX4_STEERING_MODE_B0:
724 				err = mlx4_multicast_promisc_add(mdev->dev,
725 								 priv->base_qpn,
726 								 priv->port);
727 				break;
728 
729 			case MLX4_STEERING_MODE_A0:
730 				break;
731 			}
732 			if (err)
733 				en_err(priv, "Failed entering multicast promisc mode\n");
734 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
735 		}
736 	} else {
737 		/* Disable Multicast promisc */
738 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
739 			switch (mdev->dev->caps.steering_mode) {
740 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
741 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
742 								     priv->port,
743 								     MLX4_FS_PROMISC_ALL_MULTI);
744 				break;
745 
746 			case MLX4_STEERING_MODE_B0:
747 				err = mlx4_multicast_promisc_remove(mdev->dev,
748 								    priv->base_qpn,
749 								    priv->port);
750 				break;
751 
752 			case MLX4_STEERING_MODE_A0:
753 				break;
754 			}
755 			if (err)
756 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
757 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
758 		}
759 
760 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
761 					  0, MLX4_MCAST_DISABLE);
762 		if (err)
763 			en_err(priv, "Failed disabling multicast filter\n");
764 
765 		/* Flush mcast filter and init it with broadcast address */
766 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
767 				    1, MLX4_MCAST_CONFIG);
768 
769 		/* Update multicast list - we cache all addresses so they won't
770 		 * change while HW is updated holding the command semaphor */
771 		netif_tx_lock_bh(dev);
772 		mlx4_en_cache_mclist(dev);
773 		netif_tx_unlock_bh(dev);
774 		list_for_each_entry(mclist, &priv->mc_list, list) {
775 			mcast_addr = mlx4_en_mac_to_u64(mclist->addr);
776 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
777 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
778 		}
779 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
780 					  0, MLX4_MCAST_ENABLE);
781 		if (err)
782 			en_err(priv, "Failed enabling multicast filter\n");
783 
784 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
785 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
786 			if (mclist->action == MCLIST_REM) {
787 				/* detach this address and delete from list */
788 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
789 				mc_list[5] = priv->port;
790 				err = mlx4_multicast_detach(mdev->dev,
791 							    &priv->rss_map.indir_qp,
792 							    mc_list,
793 							    MLX4_PROT_ETH,
794 							    mclist->reg_id);
795 				if (err)
796 					en_err(priv, "Fail to detach multicast address\n");
797 
798 				/* remove from list */
799 				list_del(&mclist->list);
800 				kfree(mclist);
801 			} else if (mclist->action == MCLIST_ADD) {
802 				/* attach the address */
803 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
804 				/* needed for B0 steering support */
805 				mc_list[5] = priv->port;
806 				err = mlx4_multicast_attach(mdev->dev,
807 							    &priv->rss_map.indir_qp,
808 							    mc_list,
809 							    priv->port, 0,
810 							    MLX4_PROT_ETH,
811 							    &mclist->reg_id);
812 				if (err)
813 					en_err(priv, "Fail to attach multicast address\n");
814 
815 			}
816 		}
817 	}
818 out:
819 	mutex_unlock(&mdev->state_lock);
820 }
821 
822 #ifdef CONFIG_NET_POLL_CONTROLLER
823 static void mlx4_en_netpoll(struct net_device *dev)
824 {
825 	struct mlx4_en_priv *priv = netdev_priv(dev);
826 	struct mlx4_en_cq *cq;
827 	unsigned long flags;
828 	int i;
829 
830 	for (i = 0; i < priv->rx_ring_num; i++) {
831 		cq = &priv->rx_cq[i];
832 		spin_lock_irqsave(&cq->lock, flags);
833 		napi_synchronize(&cq->napi);
834 		mlx4_en_process_rx_cq(dev, cq, 0);
835 		spin_unlock_irqrestore(&cq->lock, flags);
836 	}
837 }
838 #endif
839 
840 static void mlx4_en_tx_timeout(struct net_device *dev)
841 {
842 	struct mlx4_en_priv *priv = netdev_priv(dev);
843 	struct mlx4_en_dev *mdev = priv->mdev;
844 
845 	if (netif_msg_timer(priv))
846 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
847 
848 	priv->port_stats.tx_timeout++;
849 	en_dbg(DRV, priv, "Scheduling watchdog\n");
850 	queue_work(mdev->workqueue, &priv->watchdog_task);
851 }
852 
853 
854 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
855 {
856 	struct mlx4_en_priv *priv = netdev_priv(dev);
857 
858 	spin_lock_bh(&priv->stats_lock);
859 	memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
860 	spin_unlock_bh(&priv->stats_lock);
861 
862 	return &priv->ret_stats;
863 }
864 
865 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
866 {
867 	struct mlx4_en_cq *cq;
868 	int i;
869 
870 	/* If we haven't received a specific coalescing setting
871 	 * (module param), we set the moderation parameters as follows:
872 	 * - moder_cnt is set to the number of mtu sized packets to
873 	 *   satisfy our coelsing target.
874 	 * - moder_time is set to a fixed value.
875 	 */
876 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
877 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
878 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
879 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
880 	en_dbg(INTR, priv, "Default coalesing params for mtu:%d - "
881 			   "rx_frames:%d rx_usecs:%d\n",
882 		 priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
883 
884 	/* Setup cq moderation params */
885 	for (i = 0; i < priv->rx_ring_num; i++) {
886 		cq = &priv->rx_cq[i];
887 		cq->moder_cnt = priv->rx_frames;
888 		cq->moder_time = priv->rx_usecs;
889 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
890 		priv->last_moder_packets[i] = 0;
891 		priv->last_moder_bytes[i] = 0;
892 	}
893 
894 	for (i = 0; i < priv->tx_ring_num; i++) {
895 		cq = &priv->tx_cq[i];
896 		cq->moder_cnt = priv->tx_frames;
897 		cq->moder_time = priv->tx_usecs;
898 	}
899 
900 	/* Reset auto-moderation params */
901 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
902 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
903 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
904 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
905 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
906 	priv->adaptive_rx_coal = 1;
907 	priv->last_moder_jiffies = 0;
908 	priv->last_moder_tx_packets = 0;
909 }
910 
911 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
912 {
913 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
914 	struct mlx4_en_cq *cq;
915 	unsigned long packets;
916 	unsigned long rate;
917 	unsigned long avg_pkt_size;
918 	unsigned long rx_packets;
919 	unsigned long rx_bytes;
920 	unsigned long rx_pkt_diff;
921 	int moder_time;
922 	int ring, err;
923 
924 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
925 		return;
926 
927 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
928 		spin_lock_bh(&priv->stats_lock);
929 		rx_packets = priv->rx_ring[ring].packets;
930 		rx_bytes = priv->rx_ring[ring].bytes;
931 		spin_unlock_bh(&priv->stats_lock);
932 
933 		rx_pkt_diff = ((unsigned long) (rx_packets -
934 				priv->last_moder_packets[ring]));
935 		packets = rx_pkt_diff;
936 		rate = packets * HZ / period;
937 		avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
938 				priv->last_moder_bytes[ring])) / packets : 0;
939 
940 		/* Apply auto-moderation only when packet rate
941 		 * exceeds a rate that it matters */
942 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
943 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
944 			if (rate < priv->pkt_rate_low)
945 				moder_time = priv->rx_usecs_low;
946 			else if (rate > priv->pkt_rate_high)
947 				moder_time = priv->rx_usecs_high;
948 			else
949 				moder_time = (rate - priv->pkt_rate_low) *
950 					(priv->rx_usecs_high - priv->rx_usecs_low) /
951 					(priv->pkt_rate_high - priv->pkt_rate_low) +
952 					priv->rx_usecs_low;
953 		} else {
954 			moder_time = priv->rx_usecs_low;
955 		}
956 
957 		if (moder_time != priv->last_moder_time[ring]) {
958 			priv->last_moder_time[ring] = moder_time;
959 			cq = &priv->rx_cq[ring];
960 			cq->moder_time = moder_time;
961 			err = mlx4_en_set_cq_moder(priv, cq);
962 			if (err)
963 				en_err(priv, "Failed modifying moderation "
964 					     "for cq:%d\n", ring);
965 		}
966 		priv->last_moder_packets[ring] = rx_packets;
967 		priv->last_moder_bytes[ring] = rx_bytes;
968 	}
969 
970 	priv->last_moder_jiffies = jiffies;
971 }
972 
973 static void mlx4_en_do_get_stats(struct work_struct *work)
974 {
975 	struct delayed_work *delay = to_delayed_work(work);
976 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
977 						 stats_task);
978 	struct mlx4_en_dev *mdev = priv->mdev;
979 	int err;
980 
981 	err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
982 	if (err)
983 		en_dbg(HW, priv, "Could not update stats\n");
984 
985 	mutex_lock(&mdev->state_lock);
986 	if (mdev->device_up) {
987 		if (priv->port_up)
988 			mlx4_en_auto_moderation(priv);
989 
990 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
991 	}
992 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
993 		queue_work(mdev->workqueue, &priv->mac_task);
994 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
995 	}
996 	mutex_unlock(&mdev->state_lock);
997 }
998 
999 static void mlx4_en_linkstate(struct work_struct *work)
1000 {
1001 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1002 						 linkstate_task);
1003 	struct mlx4_en_dev *mdev = priv->mdev;
1004 	int linkstate = priv->link_state;
1005 
1006 	mutex_lock(&mdev->state_lock);
1007 	/* If observable port state changed set carrier state and
1008 	 * report to system log */
1009 	if (priv->last_link_state != linkstate) {
1010 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1011 			en_info(priv, "Link Down\n");
1012 			netif_carrier_off(priv->dev);
1013 		} else {
1014 			en_info(priv, "Link Up\n");
1015 			netif_carrier_on(priv->dev);
1016 		}
1017 	}
1018 	priv->last_link_state = linkstate;
1019 	mutex_unlock(&mdev->state_lock);
1020 }
1021 
1022 
1023 int mlx4_en_start_port(struct net_device *dev)
1024 {
1025 	struct mlx4_en_priv *priv = netdev_priv(dev);
1026 	struct mlx4_en_dev *mdev = priv->mdev;
1027 	struct mlx4_en_cq *cq;
1028 	struct mlx4_en_tx_ring *tx_ring;
1029 	int rx_index = 0;
1030 	int tx_index = 0;
1031 	int err = 0;
1032 	int i;
1033 	int j;
1034 	u8 mc_list[16] = {0};
1035 
1036 	if (priv->port_up) {
1037 		en_dbg(DRV, priv, "start port called while port already up\n");
1038 		return 0;
1039 	}
1040 
1041 	INIT_LIST_HEAD(&priv->mc_list);
1042 	INIT_LIST_HEAD(&priv->curr_list);
1043 
1044 	/* Calculate Rx buf size */
1045 	dev->mtu = min(dev->mtu, priv->max_mtu);
1046 	mlx4_en_calc_rx_buf(dev);
1047 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1048 
1049 	/* Configure rx cq's and rings */
1050 	err = mlx4_en_activate_rx_rings(priv);
1051 	if (err) {
1052 		en_err(priv, "Failed to activate RX rings\n");
1053 		return err;
1054 	}
1055 	for (i = 0; i < priv->rx_ring_num; i++) {
1056 		cq = &priv->rx_cq[i];
1057 
1058 		err = mlx4_en_activate_cq(priv, cq, i);
1059 		if (err) {
1060 			en_err(priv, "Failed activating Rx CQ\n");
1061 			goto cq_err;
1062 		}
1063 		for (j = 0; j < cq->size; j++)
1064 			cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1065 		err = mlx4_en_set_cq_moder(priv, cq);
1066 		if (err) {
1067 			en_err(priv, "Failed setting cq moderation parameters");
1068 			mlx4_en_deactivate_cq(priv, cq);
1069 			goto cq_err;
1070 		}
1071 		mlx4_en_arm_cq(priv, cq);
1072 		priv->rx_ring[i].cqn = cq->mcq.cqn;
1073 		++rx_index;
1074 	}
1075 
1076 	/* Set qp number */
1077 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1078 	err = mlx4_get_eth_qp(mdev->dev, priv->port,
1079 				priv->mac, &priv->base_qpn);
1080 	if (err) {
1081 		en_err(priv, "Failed getting eth qp\n");
1082 		goto cq_err;
1083 	}
1084 	mdev->mac_removed[priv->port] = 0;
1085 
1086 	err = mlx4_en_config_rss_steer(priv);
1087 	if (err) {
1088 		en_err(priv, "Failed configuring rss steering\n");
1089 		goto mac_err;
1090 	}
1091 
1092 	err = mlx4_en_create_drop_qp(priv);
1093 	if (err)
1094 		goto rss_err;
1095 
1096 	/* Configure tx cq's and rings */
1097 	for (i = 0; i < priv->tx_ring_num; i++) {
1098 		/* Configure cq */
1099 		cq = &priv->tx_cq[i];
1100 		err = mlx4_en_activate_cq(priv, cq, i);
1101 		if (err) {
1102 			en_err(priv, "Failed allocating Tx CQ\n");
1103 			goto tx_err;
1104 		}
1105 		err = mlx4_en_set_cq_moder(priv, cq);
1106 		if (err) {
1107 			en_err(priv, "Failed setting cq moderation parameters");
1108 			mlx4_en_deactivate_cq(priv, cq);
1109 			goto tx_err;
1110 		}
1111 		en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1112 		cq->buf->wqe_index = cpu_to_be16(0xffff);
1113 
1114 		/* Configure ring */
1115 		tx_ring = &priv->tx_ring[i];
1116 		err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1117 			i / priv->mdev->profile.num_tx_rings_p_up);
1118 		if (err) {
1119 			en_err(priv, "Failed allocating Tx ring\n");
1120 			mlx4_en_deactivate_cq(priv, cq);
1121 			goto tx_err;
1122 		}
1123 		tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1124 
1125 		/* Arm CQ for TX completions */
1126 		mlx4_en_arm_cq(priv, cq);
1127 
1128 		/* Set initial ownership of all Tx TXBBs to SW (1) */
1129 		for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1130 			*((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1131 		++tx_index;
1132 	}
1133 
1134 	/* Configure port */
1135 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1136 				    priv->rx_skb_size + ETH_FCS_LEN,
1137 				    priv->prof->tx_pause,
1138 				    priv->prof->tx_ppp,
1139 				    priv->prof->rx_pause,
1140 				    priv->prof->rx_ppp);
1141 	if (err) {
1142 		en_err(priv, "Failed setting port general configurations "
1143 			     "for port %d, with error %d\n", priv->port, err);
1144 		goto tx_err;
1145 	}
1146 	/* Set default qp number */
1147 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1148 	if (err) {
1149 		en_err(priv, "Failed setting default qp numbers\n");
1150 		goto tx_err;
1151 	}
1152 
1153 	/* Init port */
1154 	en_dbg(HW, priv, "Initializing port\n");
1155 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1156 	if (err) {
1157 		en_err(priv, "Failed Initializing port\n");
1158 		goto tx_err;
1159 	}
1160 
1161 	/* Attach rx QP to bradcast address */
1162 	memset(&mc_list[10], 0xff, ETH_ALEN);
1163 	mc_list[5] = priv->port; /* needed for B0 steering support */
1164 	if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1165 				  priv->port, 0, MLX4_PROT_ETH,
1166 				  &priv->broadcast_id))
1167 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1168 
1169 	/* Must redo promiscuous mode setup. */
1170 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1171 	if (mdev->dev->caps.steering_mode ==
1172 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1173 		mlx4_flow_steer_promisc_remove(mdev->dev,
1174 					       priv->port,
1175 					       MLX4_FS_PROMISC_UPLINK);
1176 		mlx4_flow_steer_promisc_remove(mdev->dev,
1177 					       priv->port,
1178 					       MLX4_FS_PROMISC_ALL_MULTI);
1179 	}
1180 
1181 	/* Schedule multicast task to populate multicast list */
1182 	queue_work(mdev->workqueue, &priv->mcast_task);
1183 
1184 	mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
1185 
1186 	priv->port_up = true;
1187 	netif_tx_start_all_queues(dev);
1188 	return 0;
1189 
1190 tx_err:
1191 	while (tx_index--) {
1192 		mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[tx_index]);
1193 		mlx4_en_deactivate_cq(priv, &priv->tx_cq[tx_index]);
1194 	}
1195 	mlx4_en_destroy_drop_qp(priv);
1196 rss_err:
1197 	mlx4_en_release_rss_steer(priv);
1198 mac_err:
1199 	mlx4_put_eth_qp(mdev->dev, priv->port, priv->mac, priv->base_qpn);
1200 cq_err:
1201 	while (rx_index--)
1202 		mlx4_en_deactivate_cq(priv, &priv->rx_cq[rx_index]);
1203 	for (i = 0; i < priv->rx_ring_num; i++)
1204 		mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
1205 
1206 	return err; /* need to close devices */
1207 }
1208 
1209 
1210 void mlx4_en_stop_port(struct net_device *dev)
1211 {
1212 	struct mlx4_en_priv *priv = netdev_priv(dev);
1213 	struct mlx4_en_dev *mdev = priv->mdev;
1214 	struct mlx4_en_mc_list *mclist, *tmp;
1215 	int i;
1216 	u8 mc_list[16] = {0};
1217 
1218 	if (!priv->port_up) {
1219 		en_dbg(DRV, priv, "stop port called while port already down\n");
1220 		return;
1221 	}
1222 
1223 	/* Synchronize with tx routine */
1224 	netif_tx_lock_bh(dev);
1225 	netif_tx_stop_all_queues(dev);
1226 	netif_tx_unlock_bh(dev);
1227 
1228 	/* Set port as not active */
1229 	priv->port_up = false;
1230 
1231 	/* Detach All multicasts */
1232 	memset(&mc_list[10], 0xff, ETH_ALEN);
1233 	mc_list[5] = priv->port; /* needed for B0 steering support */
1234 	mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1235 			      MLX4_PROT_ETH, priv->broadcast_id);
1236 	list_for_each_entry(mclist, &priv->curr_list, list) {
1237 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1238 		mc_list[5] = priv->port;
1239 		mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1240 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1241 	}
1242 	mlx4_en_clear_list(dev);
1243 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1244 		list_del(&mclist->list);
1245 		kfree(mclist);
1246 	}
1247 
1248 	/* Flush multicast filter */
1249 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1250 
1251 	mlx4_en_destroy_drop_qp(priv);
1252 
1253 	/* Free TX Rings */
1254 	for (i = 0; i < priv->tx_ring_num; i++) {
1255 		mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[i]);
1256 		mlx4_en_deactivate_cq(priv, &priv->tx_cq[i]);
1257 	}
1258 	msleep(10);
1259 
1260 	for (i = 0; i < priv->tx_ring_num; i++)
1261 		mlx4_en_free_tx_buf(dev, &priv->tx_ring[i]);
1262 
1263 	/* Free RSS qps */
1264 	mlx4_en_release_rss_steer(priv);
1265 
1266 	/* Unregister Mac address for the port */
1267 	mlx4_put_eth_qp(mdev->dev, priv->port, priv->mac, priv->base_qpn);
1268 	mdev->mac_removed[priv->port] = 1;
1269 
1270 	/* Free RX Rings */
1271 	for (i = 0; i < priv->rx_ring_num; i++) {
1272 		mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
1273 		while (test_bit(NAPI_STATE_SCHED, &priv->rx_cq[i].napi.state))
1274 			msleep(1);
1275 		mlx4_en_deactivate_cq(priv, &priv->rx_cq[i]);
1276 	}
1277 
1278 	/* close port*/
1279 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1280 }
1281 
1282 static void mlx4_en_restart(struct work_struct *work)
1283 {
1284 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1285 						 watchdog_task);
1286 	struct mlx4_en_dev *mdev = priv->mdev;
1287 	struct net_device *dev = priv->dev;
1288 	int i;
1289 
1290 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1291 
1292 	mutex_lock(&mdev->state_lock);
1293 	if (priv->port_up) {
1294 		mlx4_en_stop_port(dev);
1295 		for (i = 0; i < priv->tx_ring_num; i++)
1296 			netdev_tx_reset_queue(priv->tx_ring[i].tx_queue);
1297 		if (mlx4_en_start_port(dev))
1298 			en_err(priv, "Failed restarting port %d\n", priv->port);
1299 	}
1300 	mutex_unlock(&mdev->state_lock);
1301 }
1302 
1303 static void mlx4_en_clear_stats(struct net_device *dev)
1304 {
1305 	struct mlx4_en_priv *priv = netdev_priv(dev);
1306 	struct mlx4_en_dev *mdev = priv->mdev;
1307 	int i;
1308 
1309 	if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1310 		en_dbg(HW, priv, "Failed dumping statistics\n");
1311 
1312 	memset(&priv->stats, 0, sizeof(priv->stats));
1313 	memset(&priv->pstats, 0, sizeof(priv->pstats));
1314 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1315 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1316 
1317 	for (i = 0; i < priv->tx_ring_num; i++) {
1318 		priv->tx_ring[i].bytes = 0;
1319 		priv->tx_ring[i].packets = 0;
1320 		priv->tx_ring[i].tx_csum = 0;
1321 	}
1322 	for (i = 0; i < priv->rx_ring_num; i++) {
1323 		priv->rx_ring[i].bytes = 0;
1324 		priv->rx_ring[i].packets = 0;
1325 		priv->rx_ring[i].csum_ok = 0;
1326 		priv->rx_ring[i].csum_none = 0;
1327 	}
1328 }
1329 
1330 static int mlx4_en_open(struct net_device *dev)
1331 {
1332 	struct mlx4_en_priv *priv = netdev_priv(dev);
1333 	struct mlx4_en_dev *mdev = priv->mdev;
1334 	int err = 0;
1335 
1336 	mutex_lock(&mdev->state_lock);
1337 
1338 	if (!mdev->device_up) {
1339 		en_err(priv, "Cannot open - device down/disabled\n");
1340 		err = -EBUSY;
1341 		goto out;
1342 	}
1343 
1344 	/* Reset HW statistics and SW counters */
1345 	mlx4_en_clear_stats(dev);
1346 
1347 	err = mlx4_en_start_port(dev);
1348 	if (err)
1349 		en_err(priv, "Failed starting port:%d\n", priv->port);
1350 
1351 out:
1352 	mutex_unlock(&mdev->state_lock);
1353 	return err;
1354 }
1355 
1356 
1357 static int mlx4_en_close(struct net_device *dev)
1358 {
1359 	struct mlx4_en_priv *priv = netdev_priv(dev);
1360 	struct mlx4_en_dev *mdev = priv->mdev;
1361 
1362 	en_dbg(IFDOWN, priv, "Close port called\n");
1363 
1364 	mutex_lock(&mdev->state_lock);
1365 
1366 	mlx4_en_stop_port(dev);
1367 	netif_carrier_off(dev);
1368 
1369 	mutex_unlock(&mdev->state_lock);
1370 	return 0;
1371 }
1372 
1373 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1374 {
1375 	int i;
1376 
1377 #ifdef CONFIG_RFS_ACCEL
1378 	free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1379 	priv->dev->rx_cpu_rmap = NULL;
1380 #endif
1381 
1382 	for (i = 0; i < priv->tx_ring_num; i++) {
1383 		if (priv->tx_ring[i].tx_info)
1384 			mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1385 		if (priv->tx_cq[i].buf)
1386 			mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1387 	}
1388 
1389 	for (i = 0; i < priv->rx_ring_num; i++) {
1390 		if (priv->rx_ring[i].rx_info)
1391 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1392 				priv->prof->rx_ring_size, priv->stride);
1393 		if (priv->rx_cq[i].buf)
1394 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1395 	}
1396 
1397 	if (priv->base_tx_qpn) {
1398 		mlx4_qp_release_range(priv->mdev->dev, priv->base_tx_qpn, priv->tx_ring_num);
1399 		priv->base_tx_qpn = 0;
1400 	}
1401 }
1402 
1403 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1404 {
1405 	struct mlx4_en_port_profile *prof = priv->prof;
1406 	int i;
1407 	int err;
1408 
1409 	err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &priv->base_tx_qpn);
1410 	if (err) {
1411 		en_err(priv, "failed reserving range for TX rings\n");
1412 		return err;
1413 	}
1414 
1415 	/* Create tx Rings */
1416 	for (i = 0; i < priv->tx_ring_num; i++) {
1417 		if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1418 				      prof->tx_ring_size, i, TX))
1419 			goto err;
1420 
1421 		if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i], priv->base_tx_qpn + i,
1422 					   prof->tx_ring_size, TXBB_SIZE))
1423 			goto err;
1424 	}
1425 
1426 	/* Create rx Rings */
1427 	for (i = 0; i < priv->rx_ring_num; i++) {
1428 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
1429 				      prof->rx_ring_size, i, RX))
1430 			goto err;
1431 
1432 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
1433 					   prof->rx_ring_size, priv->stride))
1434 			goto err;
1435 	}
1436 
1437 #ifdef CONFIG_RFS_ACCEL
1438 	priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->rx_ring_num);
1439 	if (!priv->dev->rx_cpu_rmap)
1440 		goto err;
1441 
1442 	INIT_LIST_HEAD(&priv->filters);
1443 	spin_lock_init(&priv->filters_lock);
1444 #endif
1445 
1446 	return 0;
1447 
1448 err:
1449 	en_err(priv, "Failed to allocate NIC resources\n");
1450 	return -ENOMEM;
1451 }
1452 
1453 
1454 void mlx4_en_destroy_netdev(struct net_device *dev)
1455 {
1456 	struct mlx4_en_priv *priv = netdev_priv(dev);
1457 	struct mlx4_en_dev *mdev = priv->mdev;
1458 
1459 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
1460 
1461 	/* Unregister device - this will close the port if it was up */
1462 	if (priv->registered)
1463 		unregister_netdev(dev);
1464 
1465 	if (priv->allocated)
1466 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
1467 
1468 	cancel_delayed_work(&priv->stats_task);
1469 	/* flush any pending task for this netdev */
1470 	flush_workqueue(mdev->workqueue);
1471 
1472 	/* Detach the netdev so tasks would not attempt to access it */
1473 	mutex_lock(&mdev->state_lock);
1474 	mdev->pndev[priv->port] = NULL;
1475 	mutex_unlock(&mdev->state_lock);
1476 
1477 	mlx4_en_free_resources(priv);
1478 
1479 	kfree(priv->tx_ring);
1480 	kfree(priv->tx_cq);
1481 
1482 	free_netdev(dev);
1483 }
1484 
1485 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
1486 {
1487 	struct mlx4_en_priv *priv = netdev_priv(dev);
1488 	struct mlx4_en_dev *mdev = priv->mdev;
1489 	int err = 0;
1490 
1491 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
1492 		 dev->mtu, new_mtu);
1493 
1494 	if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
1495 		en_err(priv, "Bad MTU size:%d.\n", new_mtu);
1496 		return -EPERM;
1497 	}
1498 	dev->mtu = new_mtu;
1499 
1500 	if (netif_running(dev)) {
1501 		mutex_lock(&mdev->state_lock);
1502 		if (!mdev->device_up) {
1503 			/* NIC is probably restarting - let watchdog task reset
1504 			 * the port */
1505 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
1506 		} else {
1507 			mlx4_en_stop_port(dev);
1508 			err = mlx4_en_start_port(dev);
1509 			if (err) {
1510 				en_err(priv, "Failed restarting port:%d\n",
1511 					 priv->port);
1512 				queue_work(mdev->workqueue, &priv->watchdog_task);
1513 			}
1514 		}
1515 		mutex_unlock(&mdev->state_lock);
1516 	}
1517 	return 0;
1518 }
1519 
1520 static int mlx4_en_set_features(struct net_device *netdev,
1521 		netdev_features_t features)
1522 {
1523 	struct mlx4_en_priv *priv = netdev_priv(netdev);
1524 
1525 	if (features & NETIF_F_LOOPBACK)
1526 		priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
1527 	else
1528 		priv->ctrl_flags &=
1529 			cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
1530 
1531 	return 0;
1532 
1533 }
1534 
1535 static const struct net_device_ops mlx4_netdev_ops = {
1536 	.ndo_open		= mlx4_en_open,
1537 	.ndo_stop		= mlx4_en_close,
1538 	.ndo_start_xmit		= mlx4_en_xmit,
1539 	.ndo_select_queue	= mlx4_en_select_queue,
1540 	.ndo_get_stats		= mlx4_en_get_stats,
1541 	.ndo_set_rx_mode	= mlx4_en_set_multicast,
1542 	.ndo_set_mac_address	= mlx4_en_set_mac,
1543 	.ndo_validate_addr	= eth_validate_addr,
1544 	.ndo_change_mtu		= mlx4_en_change_mtu,
1545 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
1546 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
1547 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
1548 #ifdef CONFIG_NET_POLL_CONTROLLER
1549 	.ndo_poll_controller	= mlx4_en_netpoll,
1550 #endif
1551 	.ndo_set_features	= mlx4_en_set_features,
1552 	.ndo_setup_tc		= mlx4_en_setup_tc,
1553 #ifdef CONFIG_RFS_ACCEL
1554 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
1555 #endif
1556 };
1557 
1558 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
1559 			struct mlx4_en_port_profile *prof)
1560 {
1561 	struct net_device *dev;
1562 	struct mlx4_en_priv *priv;
1563 	int i;
1564 	int err;
1565 
1566 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
1567 	    prof->tx_ring_num, prof->rx_ring_num);
1568 	if (dev == NULL)
1569 		return -ENOMEM;
1570 
1571 	SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
1572 	dev->dev_id =  port - 1;
1573 
1574 	/*
1575 	 * Initialize driver private data
1576 	 */
1577 
1578 	priv = netdev_priv(dev);
1579 	memset(priv, 0, sizeof(struct mlx4_en_priv));
1580 	priv->dev = dev;
1581 	priv->mdev = mdev;
1582 	priv->ddev = &mdev->pdev->dev;
1583 	priv->prof = prof;
1584 	priv->port = port;
1585 	priv->port_up = false;
1586 	priv->flags = prof->flags;
1587 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
1588 			MLX4_WQE_CTRL_SOLICITED);
1589 	priv->tx_ring_num = prof->tx_ring_num;
1590 	priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring) *
1591 			priv->tx_ring_num, GFP_KERNEL);
1592 	if (!priv->tx_ring) {
1593 		err = -ENOMEM;
1594 		goto out;
1595 	}
1596 	priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq) * priv->tx_ring_num,
1597 			GFP_KERNEL);
1598 	if (!priv->tx_cq) {
1599 		err = -ENOMEM;
1600 		goto out;
1601 	}
1602 	priv->rx_ring_num = prof->rx_ring_num;
1603 	priv->mac_index = -1;
1604 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
1605 	spin_lock_init(&priv->stats_lock);
1606 	INIT_WORK(&priv->mcast_task, mlx4_en_do_set_multicast);
1607 	INIT_WORK(&priv->mac_task, mlx4_en_do_set_mac);
1608 	INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
1609 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
1610 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
1611 #ifdef CONFIG_MLX4_EN_DCB
1612 	if (!mlx4_is_slave(priv->mdev->dev))
1613 		dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
1614 #endif
1615 
1616 	/* Query for default mac and max mtu */
1617 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
1618 	priv->mac = mdev->dev->caps.def_mac[priv->port];
1619 	if (ILLEGAL_MAC(priv->mac)) {
1620 		en_err(priv, "Port: %d, invalid mac burned: 0x%llx, quiting\n",
1621 			 priv->port, priv->mac);
1622 		err = -EINVAL;
1623 		goto out;
1624 	}
1625 
1626 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
1627 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
1628 	err = mlx4_en_alloc_resources(priv);
1629 	if (err)
1630 		goto out;
1631 
1632 	/* Allocate page for receive rings */
1633 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
1634 				MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
1635 	if (err) {
1636 		en_err(priv, "Failed to allocate page for rx qps\n");
1637 		goto out;
1638 	}
1639 	priv->allocated = 1;
1640 
1641 	/*
1642 	 * Initialize netdev entry points
1643 	 */
1644 	dev->netdev_ops = &mlx4_netdev_ops;
1645 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
1646 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
1647 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
1648 
1649 	SET_ETHTOOL_OPS(dev, &mlx4_en_ethtool_ops);
1650 
1651 	/* Set defualt MAC */
1652 	dev->addr_len = ETH_ALEN;
1653 	for (i = 0; i < ETH_ALEN; i++) {
1654 		dev->dev_addr[ETH_ALEN - 1 - i] = (u8) (priv->mac >> (8 * i));
1655 		dev->perm_addr[ETH_ALEN - 1 - i] = (u8) (priv->mac >> (8 * i));
1656 	}
1657 
1658 	/*
1659 	 * Set driver features
1660 	 */
1661 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1662 	if (mdev->LSO_support)
1663 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1664 
1665 	dev->vlan_features = dev->hw_features;
1666 
1667 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
1668 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
1669 			NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
1670 			NETIF_F_HW_VLAN_FILTER;
1671 	dev->hw_features |= NETIF_F_LOOPBACK;
1672 
1673 	if (mdev->dev->caps.steering_mode ==
1674 	    MLX4_STEERING_MODE_DEVICE_MANAGED)
1675 		dev->hw_features |= NETIF_F_NTUPLE;
1676 
1677 	mdev->pndev[port] = dev;
1678 
1679 	netif_carrier_off(dev);
1680 	err = register_netdev(dev);
1681 	if (err) {
1682 		en_err(priv, "Netdev registration failed for port %d\n", port);
1683 		goto out;
1684 	}
1685 	priv->registered = 1;
1686 
1687 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
1688 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
1689 
1690 	/* Configure port */
1691 	mlx4_en_calc_rx_buf(dev);
1692 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1693 				    priv->rx_skb_size + ETH_FCS_LEN,
1694 				    prof->tx_pause, prof->tx_ppp,
1695 				    prof->rx_pause, prof->rx_ppp);
1696 	if (err) {
1697 		en_err(priv, "Failed setting port general configurations "
1698 		       "for port %d, with error %d\n", priv->port, err);
1699 		goto out;
1700 	}
1701 
1702 	/* Init port */
1703 	en_warn(priv, "Initializing port\n");
1704 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1705 	if (err) {
1706 		en_err(priv, "Failed Initializing port\n");
1707 		goto out;
1708 	}
1709 	mlx4_en_set_default_moderation(priv);
1710 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1711 	return 0;
1712 
1713 out:
1714 	mlx4_en_destroy_netdev(dev);
1715 	return err;
1716 }
1717 
1718