• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 
34 #include <linux/bpf.h>
35 #include <linux/etherdevice.h>
36 #include <linux/tcp.h>
37 #include <linux/if_vlan.h>
38 #include <linux/delay.h>
39 #include <linux/slab.h>
40 #include <linux/hash.h>
41 #include <net/ip.h>
42 #include <net/vxlan.h>
43 #include <net/devlink.h>
44 
45 #include <linux/mlx4/driver.h>
46 #include <linux/mlx4/device.h>
47 #include <linux/mlx4/cmd.h>
48 #include <linux/mlx4/cq.h>
49 
50 #include "mlx4_en.h"
51 #include "en_port.h"
52 
53 #define MLX4_EN_MAX_XDP_MTU ((int)(PAGE_SIZE - ETH_HLEN - (2 * VLAN_HLEN) - \
54 				XDP_PACKET_HEADROOM -			    \
55 				SKB_DATA_ALIGN(sizeof(struct skb_shared_info))))
56 
mlx4_en_setup_tc(struct net_device * dev,u8 up)57 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
58 {
59 	struct mlx4_en_priv *priv = netdev_priv(dev);
60 	int i;
61 	unsigned int offset = 0;
62 
63 	if (up && up != MLX4_EN_NUM_UP_HIGH)
64 		return -EINVAL;
65 
66 	netdev_set_num_tc(dev, up);
67 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
68 	/* Partition Tx queues evenly amongst UP's */
69 	for (i = 0; i < up; i++) {
70 		netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
71 		offset += priv->num_tx_rings_p_up;
72 	}
73 
74 #ifdef CONFIG_MLX4_EN_DCB
75 	if (!mlx4_is_slave(priv->mdev->dev)) {
76 		if (up) {
77 			if (priv->dcbx_cap)
78 				priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
79 		} else {
80 			priv->flags &= ~MLX4_EN_FLAG_DCB_ENABLED;
81 			priv->cee_config.pfc_state = false;
82 		}
83 	}
84 #endif /* CONFIG_MLX4_EN_DCB */
85 
86 	return 0;
87 }
88 
mlx4_en_alloc_tx_queue_per_tc(struct net_device * dev,u8 tc)89 int mlx4_en_alloc_tx_queue_per_tc(struct net_device *dev, u8 tc)
90 {
91 	struct mlx4_en_priv *priv = netdev_priv(dev);
92 	struct mlx4_en_dev *mdev = priv->mdev;
93 	struct mlx4_en_port_profile new_prof;
94 	struct mlx4_en_priv *tmp;
95 	int total_count;
96 	int port_up = 0;
97 	int err = 0;
98 
99 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
100 	if (!tmp)
101 		return -ENOMEM;
102 
103 	mutex_lock(&mdev->state_lock);
104 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
105 	new_prof.num_up = (tc == 0) ? MLX4_EN_NUM_UP_LOW :
106 				      MLX4_EN_NUM_UP_HIGH;
107 	new_prof.tx_ring_num[TX] = new_prof.num_tx_rings_p_up *
108 				   new_prof.num_up;
109 	total_count = new_prof.tx_ring_num[TX] + new_prof.tx_ring_num[TX_XDP];
110 	if (total_count > MAX_TX_RINGS) {
111 		err = -EINVAL;
112 		en_err(priv,
113 		       "Total number of TX and XDP rings (%d) exceeds the maximum supported (%d)\n",
114 		       total_count, MAX_TX_RINGS);
115 		goto out;
116 	}
117 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
118 	if (err)
119 		goto out;
120 
121 	if (priv->port_up) {
122 		port_up = 1;
123 		mlx4_en_stop_port(dev, 1);
124 	}
125 
126 	mlx4_en_safe_replace_resources(priv, tmp);
127 	if (port_up) {
128 		err = mlx4_en_start_port(dev);
129 		if (err) {
130 			en_err(priv, "Failed starting port for setup TC\n");
131 			goto out;
132 		}
133 	}
134 
135 	err = mlx4_en_setup_tc(dev, tc);
136 out:
137 	mutex_unlock(&mdev->state_lock);
138 	kfree(tmp);
139 	return err;
140 }
141 
__mlx4_en_setup_tc(struct net_device * dev,enum tc_setup_type type,void * type_data)142 static int __mlx4_en_setup_tc(struct net_device *dev, enum tc_setup_type type,
143 			      void *type_data)
144 {
145 	struct tc_mqprio_qopt *mqprio = type_data;
146 
147 	if (type != TC_SETUP_QDISC_MQPRIO)
148 		return -EOPNOTSUPP;
149 
150 	if (mqprio->num_tc && mqprio->num_tc != MLX4_EN_NUM_UP_HIGH)
151 		return -EINVAL;
152 
153 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
154 
155 	return mlx4_en_alloc_tx_queue_per_tc(dev, mqprio->num_tc);
156 }
157 
158 #ifdef CONFIG_RFS_ACCEL
159 
160 struct mlx4_en_filter {
161 	struct list_head next;
162 	struct work_struct work;
163 
164 	u8     ip_proto;
165 	__be32 src_ip;
166 	__be32 dst_ip;
167 	__be16 src_port;
168 	__be16 dst_port;
169 
170 	int rxq_index;
171 	struct mlx4_en_priv *priv;
172 	u32 flow_id;			/* RFS infrastructure id */
173 	int id;				/* mlx4_en driver id */
174 	u64 reg_id;			/* Flow steering API id */
175 	u8 activated;			/* Used to prevent expiry before filter
176 					 * is attached
177 					 */
178 	struct hlist_node filter_chain;
179 };
180 
181 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
182 
mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)183 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
184 {
185 	switch (ip_proto) {
186 	case IPPROTO_UDP:
187 		return MLX4_NET_TRANS_RULE_ID_UDP;
188 	case IPPROTO_TCP:
189 		return MLX4_NET_TRANS_RULE_ID_TCP;
190 	default:
191 		return MLX4_NET_TRANS_RULE_NUM;
192 	}
193 };
194 
195 /* Must not acquire state_lock, as its corresponding work_sync
196  * is done under it.
197  */
mlx4_en_filter_work(struct work_struct * work)198 static void mlx4_en_filter_work(struct work_struct *work)
199 {
200 	struct mlx4_en_filter *filter = container_of(work,
201 						     struct mlx4_en_filter,
202 						     work);
203 	struct mlx4_en_priv *priv = filter->priv;
204 	struct mlx4_spec_list spec_tcp_udp = {
205 		.id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
206 		{
207 			.tcp_udp = {
208 				.dst_port = filter->dst_port,
209 				.dst_port_msk = (__force __be16)-1,
210 				.src_port = filter->src_port,
211 				.src_port_msk = (__force __be16)-1,
212 			},
213 		},
214 	};
215 	struct mlx4_spec_list spec_ip = {
216 		.id = MLX4_NET_TRANS_RULE_ID_IPV4,
217 		{
218 			.ipv4 = {
219 				.dst_ip = filter->dst_ip,
220 				.dst_ip_msk = (__force __be32)-1,
221 				.src_ip = filter->src_ip,
222 				.src_ip_msk = (__force __be32)-1,
223 			},
224 		},
225 	};
226 	struct mlx4_spec_list spec_eth = {
227 		.id = MLX4_NET_TRANS_RULE_ID_ETH,
228 	};
229 	struct mlx4_net_trans_rule rule = {
230 		.list = LIST_HEAD_INIT(rule.list),
231 		.queue_mode = MLX4_NET_TRANS_Q_LIFO,
232 		.exclusive = 1,
233 		.allow_loopback = 1,
234 		.promisc_mode = MLX4_FS_REGULAR,
235 		.port = priv->port,
236 		.priority = MLX4_DOMAIN_RFS,
237 	};
238 	int rc;
239 	__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
240 
241 	if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
242 		en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
243 			filter->ip_proto);
244 		goto ignore;
245 	}
246 	list_add_tail(&spec_eth.list, &rule.list);
247 	list_add_tail(&spec_ip.list, &rule.list);
248 	list_add_tail(&spec_tcp_udp.list, &rule.list);
249 
250 	rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
251 	memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
252 	memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
253 
254 	filter->activated = 0;
255 
256 	if (filter->reg_id) {
257 		rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
258 		if (rc && rc != -ENOENT)
259 			en_err(priv, "Error detaching flow. rc = %d\n", rc);
260 	}
261 
262 	rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
263 	if (rc)
264 		en_err(priv, "Error attaching flow. err = %d\n", rc);
265 
266 ignore:
267 	mlx4_en_filter_rfs_expire(priv);
268 
269 	filter->activated = 1;
270 }
271 
272 static inline struct hlist_head *
filter_hash_bucket(struct mlx4_en_priv * priv,__be32 src_ip,__be32 dst_ip,__be16 src_port,__be16 dst_port)273 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
274 		   __be16 src_port, __be16 dst_port)
275 {
276 	unsigned long l;
277 	int bucket_idx;
278 
279 	l = (__force unsigned long)src_port |
280 	    ((__force unsigned long)dst_port << 2);
281 	l ^= (__force unsigned long)(src_ip ^ dst_ip);
282 
283 	bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
284 
285 	return &priv->filter_hash[bucket_idx];
286 }
287 
288 static struct mlx4_en_filter *
mlx4_en_filter_alloc(struct mlx4_en_priv * priv,int rxq_index,__be32 src_ip,__be32 dst_ip,u8 ip_proto,__be16 src_port,__be16 dst_port,u32 flow_id)289 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
290 		     __be32 dst_ip, u8 ip_proto, __be16 src_port,
291 		     __be16 dst_port, u32 flow_id)
292 {
293 	struct mlx4_en_filter *filter = NULL;
294 
295 	filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
296 	if (!filter)
297 		return NULL;
298 
299 	filter->priv = priv;
300 	filter->rxq_index = rxq_index;
301 	INIT_WORK(&filter->work, mlx4_en_filter_work);
302 
303 	filter->src_ip = src_ip;
304 	filter->dst_ip = dst_ip;
305 	filter->ip_proto = ip_proto;
306 	filter->src_port = src_port;
307 	filter->dst_port = dst_port;
308 
309 	filter->flow_id = flow_id;
310 
311 	filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
312 
313 	list_add_tail(&filter->next, &priv->filters);
314 	hlist_add_head(&filter->filter_chain,
315 		       filter_hash_bucket(priv, src_ip, dst_ip, src_port,
316 					  dst_port));
317 
318 	return filter;
319 }
320 
mlx4_en_filter_free(struct mlx4_en_filter * filter)321 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
322 {
323 	struct mlx4_en_priv *priv = filter->priv;
324 	int rc;
325 
326 	list_del(&filter->next);
327 
328 	rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
329 	if (rc && rc != -ENOENT)
330 		en_err(priv, "Error detaching flow. rc = %d\n", rc);
331 
332 	kfree(filter);
333 }
334 
335 static inline struct mlx4_en_filter *
mlx4_en_filter_find(struct mlx4_en_priv * priv,__be32 src_ip,__be32 dst_ip,u8 ip_proto,__be16 src_port,__be16 dst_port)336 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
337 		    u8 ip_proto, __be16 src_port, __be16 dst_port)
338 {
339 	struct mlx4_en_filter *filter;
340 	struct mlx4_en_filter *ret = NULL;
341 
342 	hlist_for_each_entry(filter,
343 			     filter_hash_bucket(priv, src_ip, dst_ip,
344 						src_port, dst_port),
345 			     filter_chain) {
346 		if (filter->src_ip == src_ip &&
347 		    filter->dst_ip == dst_ip &&
348 		    filter->ip_proto == ip_proto &&
349 		    filter->src_port == src_port &&
350 		    filter->dst_port == dst_port) {
351 			ret = filter;
352 			break;
353 		}
354 	}
355 
356 	return ret;
357 }
358 
359 static int
mlx4_en_filter_rfs(struct net_device * net_dev,const struct sk_buff * skb,u16 rxq_index,u32 flow_id)360 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
361 		   u16 rxq_index, u32 flow_id)
362 {
363 	struct mlx4_en_priv *priv = netdev_priv(net_dev);
364 	struct mlx4_en_filter *filter;
365 	const struct iphdr *ip;
366 	const __be16 *ports;
367 	u8 ip_proto;
368 	__be32 src_ip;
369 	__be32 dst_ip;
370 	__be16 src_port;
371 	__be16 dst_port;
372 	int nhoff = skb_network_offset(skb);
373 	int ret = 0;
374 
375 	if (skb->encapsulation)
376 		return -EPROTONOSUPPORT;
377 
378 	if (skb->protocol != htons(ETH_P_IP))
379 		return -EPROTONOSUPPORT;
380 
381 	ip = (const struct iphdr *)(skb->data + nhoff);
382 	if (ip_is_fragment(ip))
383 		return -EPROTONOSUPPORT;
384 
385 	if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
386 		return -EPROTONOSUPPORT;
387 	ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
388 
389 	ip_proto = ip->protocol;
390 	src_ip = ip->saddr;
391 	dst_ip = ip->daddr;
392 	src_port = ports[0];
393 	dst_port = ports[1];
394 
395 	spin_lock_bh(&priv->filters_lock);
396 	filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
397 				     src_port, dst_port);
398 	if (filter) {
399 		if (filter->rxq_index == rxq_index)
400 			goto out;
401 
402 		filter->rxq_index = rxq_index;
403 	} else {
404 		filter = mlx4_en_filter_alloc(priv, rxq_index,
405 					      src_ip, dst_ip, ip_proto,
406 					      src_port, dst_port, flow_id);
407 		if (!filter) {
408 			ret = -ENOMEM;
409 			goto err;
410 		}
411 	}
412 
413 	queue_work(priv->mdev->workqueue, &filter->work);
414 
415 out:
416 	ret = filter->id;
417 err:
418 	spin_unlock_bh(&priv->filters_lock);
419 
420 	return ret;
421 }
422 
mlx4_en_cleanup_filters(struct mlx4_en_priv * priv)423 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
424 {
425 	struct mlx4_en_filter *filter, *tmp;
426 	LIST_HEAD(del_list);
427 
428 	spin_lock_bh(&priv->filters_lock);
429 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
430 		list_move(&filter->next, &del_list);
431 		hlist_del(&filter->filter_chain);
432 	}
433 	spin_unlock_bh(&priv->filters_lock);
434 
435 	list_for_each_entry_safe(filter, tmp, &del_list, next) {
436 		cancel_work_sync(&filter->work);
437 		mlx4_en_filter_free(filter);
438 	}
439 }
440 
mlx4_en_filter_rfs_expire(struct mlx4_en_priv * priv)441 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
442 {
443 	struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
444 	LIST_HEAD(del_list);
445 	int i = 0;
446 
447 	spin_lock_bh(&priv->filters_lock);
448 	list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
449 		if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
450 			break;
451 
452 		if (filter->activated &&
453 		    !work_pending(&filter->work) &&
454 		    rps_may_expire_flow(priv->dev,
455 					filter->rxq_index, filter->flow_id,
456 					filter->id)) {
457 			list_move(&filter->next, &del_list);
458 			hlist_del(&filter->filter_chain);
459 		} else
460 			last_filter = filter;
461 
462 		i++;
463 	}
464 
465 	if (last_filter && (&last_filter->next != priv->filters.next))
466 		list_move(&priv->filters, &last_filter->next);
467 
468 	spin_unlock_bh(&priv->filters_lock);
469 
470 	list_for_each_entry_safe(filter, tmp, &del_list, next)
471 		mlx4_en_filter_free(filter);
472 }
473 #endif
474 
mlx4_en_vlan_rx_add_vid(struct net_device * dev,__be16 proto,u16 vid)475 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
476 				   __be16 proto, u16 vid)
477 {
478 	struct mlx4_en_priv *priv = netdev_priv(dev);
479 	struct mlx4_en_dev *mdev = priv->mdev;
480 	int err;
481 	int idx;
482 
483 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
484 
485 	set_bit(vid, priv->active_vlans);
486 
487 	/* Add VID to port VLAN filter */
488 	mutex_lock(&mdev->state_lock);
489 	if (mdev->device_up && priv->port_up) {
490 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
491 		if (err) {
492 			en_err(priv, "Failed configuring VLAN filter\n");
493 			goto out;
494 		}
495 	}
496 	err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
497 	if (err)
498 		en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
499 
500 out:
501 	mutex_unlock(&mdev->state_lock);
502 	return err;
503 }
504 
mlx4_en_vlan_rx_kill_vid(struct net_device * dev,__be16 proto,u16 vid)505 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
506 				    __be16 proto, u16 vid)
507 {
508 	struct mlx4_en_priv *priv = netdev_priv(dev);
509 	struct mlx4_en_dev *mdev = priv->mdev;
510 	int err = 0;
511 
512 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
513 
514 	clear_bit(vid, priv->active_vlans);
515 
516 	/* Remove VID from port VLAN filter */
517 	mutex_lock(&mdev->state_lock);
518 	mlx4_unregister_vlan(mdev->dev, priv->port, vid);
519 
520 	if (mdev->device_up && priv->port_up) {
521 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
522 		if (err)
523 			en_err(priv, "Failed configuring VLAN filter\n");
524 	}
525 	mutex_unlock(&mdev->state_lock);
526 
527 	return err;
528 }
529 
mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN+2],u64 src_mac)530 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
531 {
532 	int i;
533 	for (i = ETH_ALEN - 1; i >= 0; --i) {
534 		dst_mac[i] = src_mac & 0xff;
535 		src_mac >>= 8;
536 	}
537 	memset(&dst_mac[ETH_ALEN], 0, 2);
538 }
539 
540 
mlx4_en_tunnel_steer_add(struct mlx4_en_priv * priv,unsigned char * addr,int qpn,u64 * reg_id)541 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
542 				    int qpn, u64 *reg_id)
543 {
544 	int err;
545 
546 	if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
547 	    priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
548 		return 0; /* do nothing */
549 
550 	err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
551 				    MLX4_DOMAIN_NIC, reg_id);
552 	if (err) {
553 		en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
554 		return err;
555 	}
556 	en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
557 	return 0;
558 }
559 
560 
mlx4_en_uc_steer_add(struct mlx4_en_priv * priv,unsigned char * mac,int * qpn,u64 * reg_id)561 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
562 				unsigned char *mac, int *qpn, u64 *reg_id)
563 {
564 	struct mlx4_en_dev *mdev = priv->mdev;
565 	struct mlx4_dev *dev = mdev->dev;
566 	int err;
567 
568 	switch (dev->caps.steering_mode) {
569 	case MLX4_STEERING_MODE_B0: {
570 		struct mlx4_qp qp;
571 		u8 gid[16] = {0};
572 
573 		qp.qpn = *qpn;
574 		memcpy(&gid[10], mac, ETH_ALEN);
575 		gid[5] = priv->port;
576 
577 		err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
578 		break;
579 	}
580 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
581 		struct mlx4_spec_list spec_eth = { {NULL} };
582 		__be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
583 
584 		struct mlx4_net_trans_rule rule = {
585 			.queue_mode = MLX4_NET_TRANS_Q_FIFO,
586 			.exclusive = 0,
587 			.allow_loopback = 1,
588 			.promisc_mode = MLX4_FS_REGULAR,
589 			.priority = MLX4_DOMAIN_NIC,
590 		};
591 
592 		rule.port = priv->port;
593 		rule.qpn = *qpn;
594 		INIT_LIST_HEAD(&rule.list);
595 
596 		spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
597 		memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
598 		memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
599 		list_add_tail(&spec_eth.list, &rule.list);
600 
601 		err = mlx4_flow_attach(dev, &rule, reg_id);
602 		break;
603 	}
604 	default:
605 		return -EINVAL;
606 	}
607 	if (err)
608 		en_warn(priv, "Failed Attaching Unicast\n");
609 
610 	return err;
611 }
612 
mlx4_en_uc_steer_release(struct mlx4_en_priv * priv,unsigned char * mac,int qpn,u64 reg_id)613 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
614 				     unsigned char *mac, int qpn, u64 reg_id)
615 {
616 	struct mlx4_en_dev *mdev = priv->mdev;
617 	struct mlx4_dev *dev = mdev->dev;
618 
619 	switch (dev->caps.steering_mode) {
620 	case MLX4_STEERING_MODE_B0: {
621 		struct mlx4_qp qp;
622 		u8 gid[16] = {0};
623 
624 		qp.qpn = qpn;
625 		memcpy(&gid[10], mac, ETH_ALEN);
626 		gid[5] = priv->port;
627 
628 		mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
629 		break;
630 	}
631 	case MLX4_STEERING_MODE_DEVICE_MANAGED: {
632 		mlx4_flow_detach(dev, reg_id);
633 		break;
634 	}
635 	default:
636 		en_err(priv, "Invalid steering mode.\n");
637 	}
638 }
639 
mlx4_en_get_qp(struct mlx4_en_priv * priv)640 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
641 {
642 	struct mlx4_en_dev *mdev = priv->mdev;
643 	struct mlx4_dev *dev = mdev->dev;
644 	int index = 0;
645 	int err = 0;
646 	int *qpn = &priv->base_qpn;
647 	u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
648 
649 	en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
650 	       priv->dev->dev_addr);
651 	index = mlx4_register_mac(dev, priv->port, mac);
652 	if (index < 0) {
653 		err = index;
654 		en_err(priv, "Failed adding MAC: %pM\n",
655 		       priv->dev->dev_addr);
656 		return err;
657 	}
658 
659 	en_info(priv, "Steering Mode %d\n", dev->caps.steering_mode);
660 
661 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
662 		int base_qpn = mlx4_get_base_qpn(dev, priv->port);
663 		*qpn = base_qpn + index;
664 		return 0;
665 	}
666 
667 	err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP,
668 				    MLX4_RES_USAGE_DRIVER);
669 	en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
670 	if (err) {
671 		en_err(priv, "Failed to reserve qp for mac registration\n");
672 		mlx4_unregister_mac(dev, priv->port, mac);
673 		return err;
674 	}
675 
676 	return 0;
677 }
678 
mlx4_en_put_qp(struct mlx4_en_priv * priv)679 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
680 {
681 	struct mlx4_en_dev *mdev = priv->mdev;
682 	struct mlx4_dev *dev = mdev->dev;
683 	int qpn = priv->base_qpn;
684 
685 	if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
686 		u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
687 		en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
688 		       priv->dev->dev_addr);
689 		mlx4_unregister_mac(dev, priv->port, mac);
690 	} else {
691 		en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
692 		       priv->port, qpn);
693 		mlx4_qp_release_range(dev, qpn, 1);
694 		priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
695 	}
696 }
697 
mlx4_en_replace_mac(struct mlx4_en_priv * priv,int qpn,unsigned char * new_mac,unsigned char * prev_mac)698 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
699 			       unsigned char *new_mac, unsigned char *prev_mac)
700 {
701 	struct mlx4_en_dev *mdev = priv->mdev;
702 	struct mlx4_dev *dev = mdev->dev;
703 	int err = 0;
704 	u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
705 
706 	if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
707 		struct hlist_head *bucket;
708 		unsigned int mac_hash;
709 		struct mlx4_mac_entry *entry;
710 		struct hlist_node *tmp;
711 		u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
712 
713 		bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
714 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
715 			if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
716 				mlx4_en_uc_steer_release(priv, entry->mac,
717 							 qpn, entry->reg_id);
718 				mlx4_unregister_mac(dev, priv->port,
719 						    prev_mac_u64);
720 				hlist_del_rcu(&entry->hlist);
721 				synchronize_rcu();
722 				memcpy(entry->mac, new_mac, ETH_ALEN);
723 				entry->reg_id = 0;
724 				mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
725 				hlist_add_head_rcu(&entry->hlist,
726 						   &priv->mac_hash[mac_hash]);
727 				mlx4_register_mac(dev, priv->port, new_mac_u64);
728 				err = mlx4_en_uc_steer_add(priv, new_mac,
729 							   &qpn,
730 							   &entry->reg_id);
731 				if (err)
732 					return err;
733 				if (priv->tunnel_reg_id) {
734 					mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
735 					priv->tunnel_reg_id = 0;
736 				}
737 				err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
738 							       &priv->tunnel_reg_id);
739 				return err;
740 			}
741 		}
742 		return -EINVAL;
743 	}
744 
745 	return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
746 }
747 
mlx4_en_update_user_mac(struct mlx4_en_priv * priv,unsigned char new_mac[ETH_ALEN+2])748 static void mlx4_en_update_user_mac(struct mlx4_en_priv *priv,
749 				    unsigned char new_mac[ETH_ALEN + 2])
750 {
751 	struct mlx4_en_dev *mdev = priv->mdev;
752 	int err;
753 
754 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_USER_MAC_EN))
755 		return;
756 
757 	err = mlx4_SET_PORT_user_mac(mdev->dev, priv->port, new_mac);
758 	if (err)
759 		en_err(priv, "Failed to pass user MAC(%pM) to Firmware for port %d, with error %d\n",
760 		       new_mac, priv->port, err);
761 }
762 
mlx4_en_do_set_mac(struct mlx4_en_priv * priv,unsigned char new_mac[ETH_ALEN+2])763 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
764 			      unsigned char new_mac[ETH_ALEN + 2])
765 {
766 	int err = 0;
767 
768 	if (priv->port_up) {
769 		/* Remove old MAC and insert the new one */
770 		err = mlx4_en_replace_mac(priv, priv->base_qpn,
771 					  new_mac, priv->current_mac);
772 		if (err)
773 			en_err(priv, "Failed changing HW MAC address\n");
774 	} else
775 		en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
776 
777 	if (!err)
778 		memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
779 
780 	return err;
781 }
782 
mlx4_en_set_mac(struct net_device * dev,void * addr)783 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
784 {
785 	struct mlx4_en_priv *priv = netdev_priv(dev);
786 	struct mlx4_en_dev *mdev = priv->mdev;
787 	struct sockaddr *saddr = addr;
788 	unsigned char new_mac[ETH_ALEN + 2];
789 	int err;
790 
791 	if (!is_valid_ether_addr(saddr->sa_data))
792 		return -EADDRNOTAVAIL;
793 
794 	mutex_lock(&mdev->state_lock);
795 	memcpy(new_mac, saddr->sa_data, ETH_ALEN);
796 	err = mlx4_en_do_set_mac(priv, new_mac);
797 	if (err)
798 		goto out;
799 
800 	memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
801 	mlx4_en_update_user_mac(priv, new_mac);
802 out:
803 	mutex_unlock(&mdev->state_lock);
804 
805 	return err;
806 }
807 
mlx4_en_clear_list(struct net_device * dev)808 static void mlx4_en_clear_list(struct net_device *dev)
809 {
810 	struct mlx4_en_priv *priv = netdev_priv(dev);
811 	struct mlx4_en_mc_list *tmp, *mc_to_del;
812 
813 	list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
814 		list_del(&mc_to_del->list);
815 		kfree(mc_to_del);
816 	}
817 }
818 
mlx4_en_cache_mclist(struct net_device * dev)819 static void mlx4_en_cache_mclist(struct net_device *dev)
820 {
821 	struct mlx4_en_priv *priv = netdev_priv(dev);
822 	struct netdev_hw_addr *ha;
823 	struct mlx4_en_mc_list *tmp;
824 
825 	mlx4_en_clear_list(dev);
826 	netdev_for_each_mc_addr(ha, dev) {
827 		tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
828 		if (!tmp) {
829 			mlx4_en_clear_list(dev);
830 			return;
831 		}
832 		memcpy(tmp->addr, ha->addr, ETH_ALEN);
833 		list_add_tail(&tmp->list, &priv->mc_list);
834 	}
835 }
836 
update_mclist_flags(struct mlx4_en_priv * priv,struct list_head * dst,struct list_head * src)837 static void update_mclist_flags(struct mlx4_en_priv *priv,
838 				struct list_head *dst,
839 				struct list_head *src)
840 {
841 	struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
842 	bool found;
843 
844 	/* Find all the entries that should be removed from dst,
845 	 * These are the entries that are not found in src
846 	 */
847 	list_for_each_entry(dst_tmp, dst, list) {
848 		found = false;
849 		list_for_each_entry(src_tmp, src, list) {
850 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
851 				found = true;
852 				break;
853 			}
854 		}
855 		if (!found)
856 			dst_tmp->action = MCLIST_REM;
857 	}
858 
859 	/* Add entries that exist in src but not in dst
860 	 * mark them as need to add
861 	 */
862 	list_for_each_entry(src_tmp, src, list) {
863 		found = false;
864 		list_for_each_entry(dst_tmp, dst, list) {
865 			if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
866 				dst_tmp->action = MCLIST_NONE;
867 				found = true;
868 				break;
869 			}
870 		}
871 		if (!found) {
872 			new_mc = kmemdup(src_tmp,
873 					 sizeof(struct mlx4_en_mc_list),
874 					 GFP_KERNEL);
875 			if (!new_mc)
876 				return;
877 
878 			new_mc->action = MCLIST_ADD;
879 			list_add_tail(&new_mc->list, dst);
880 		}
881 	}
882 }
883 
mlx4_en_set_rx_mode(struct net_device * dev)884 static void mlx4_en_set_rx_mode(struct net_device *dev)
885 {
886 	struct mlx4_en_priv *priv = netdev_priv(dev);
887 
888 	if (!priv->port_up)
889 		return;
890 
891 	queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
892 }
893 
mlx4_en_set_promisc_mode(struct mlx4_en_priv * priv,struct mlx4_en_dev * mdev)894 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
895 				     struct mlx4_en_dev *mdev)
896 {
897 	int err = 0;
898 
899 	if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
900 		if (netif_msg_rx_status(priv))
901 			en_warn(priv, "Entering promiscuous mode\n");
902 		priv->flags |= MLX4_EN_FLAG_PROMISC;
903 
904 		/* Enable promiscouos mode */
905 		switch (mdev->dev->caps.steering_mode) {
906 		case MLX4_STEERING_MODE_DEVICE_MANAGED:
907 			err = mlx4_flow_steer_promisc_add(mdev->dev,
908 							  priv->port,
909 							  priv->base_qpn,
910 							  MLX4_FS_ALL_DEFAULT);
911 			if (err)
912 				en_err(priv, "Failed enabling promiscuous mode\n");
913 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
914 			break;
915 
916 		case MLX4_STEERING_MODE_B0:
917 			err = mlx4_unicast_promisc_add(mdev->dev,
918 						       priv->base_qpn,
919 						       priv->port);
920 			if (err)
921 				en_err(priv, "Failed enabling unicast promiscuous mode\n");
922 
923 			/* Add the default qp number as multicast
924 			 * promisc
925 			 */
926 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
927 				err = mlx4_multicast_promisc_add(mdev->dev,
928 								 priv->base_qpn,
929 								 priv->port);
930 				if (err)
931 					en_err(priv, "Failed enabling multicast promiscuous mode\n");
932 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
933 			}
934 			break;
935 
936 		case MLX4_STEERING_MODE_A0:
937 			err = mlx4_SET_PORT_qpn_calc(mdev->dev,
938 						     priv->port,
939 						     priv->base_qpn,
940 						     1);
941 			if (err)
942 				en_err(priv, "Failed enabling promiscuous mode\n");
943 			break;
944 		}
945 
946 		/* Disable port multicast filter (unconditionally) */
947 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
948 					  0, MLX4_MCAST_DISABLE);
949 		if (err)
950 			en_err(priv, "Failed disabling multicast filter\n");
951 	}
952 }
953 
mlx4_en_clear_promisc_mode(struct mlx4_en_priv * priv,struct mlx4_en_dev * mdev)954 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
955 				       struct mlx4_en_dev *mdev)
956 {
957 	int err = 0;
958 
959 	if (netif_msg_rx_status(priv))
960 		en_warn(priv, "Leaving promiscuous mode\n");
961 	priv->flags &= ~MLX4_EN_FLAG_PROMISC;
962 
963 	/* Disable promiscouos mode */
964 	switch (mdev->dev->caps.steering_mode) {
965 	case MLX4_STEERING_MODE_DEVICE_MANAGED:
966 		err = mlx4_flow_steer_promisc_remove(mdev->dev,
967 						     priv->port,
968 						     MLX4_FS_ALL_DEFAULT);
969 		if (err)
970 			en_err(priv, "Failed disabling promiscuous mode\n");
971 		priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
972 		break;
973 
974 	case MLX4_STEERING_MODE_B0:
975 		err = mlx4_unicast_promisc_remove(mdev->dev,
976 						  priv->base_qpn,
977 						  priv->port);
978 		if (err)
979 			en_err(priv, "Failed disabling unicast promiscuous mode\n");
980 		/* Disable Multicast promisc */
981 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
982 			err = mlx4_multicast_promisc_remove(mdev->dev,
983 							    priv->base_qpn,
984 							    priv->port);
985 			if (err)
986 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
987 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
988 		}
989 		break;
990 
991 	case MLX4_STEERING_MODE_A0:
992 		err = mlx4_SET_PORT_qpn_calc(mdev->dev,
993 					     priv->port,
994 					     priv->base_qpn, 0);
995 		if (err)
996 			en_err(priv, "Failed disabling promiscuous mode\n");
997 		break;
998 	}
999 }
1000 
mlx4_en_do_multicast(struct mlx4_en_priv * priv,struct net_device * dev,struct mlx4_en_dev * mdev)1001 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
1002 				 struct net_device *dev,
1003 				 struct mlx4_en_dev *mdev)
1004 {
1005 	struct mlx4_en_mc_list *mclist, *tmp;
1006 	u64 mcast_addr = 0;
1007 	u8 mc_list[16] = {0};
1008 	int err = 0;
1009 
1010 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
1011 	if (dev->flags & IFF_ALLMULTI) {
1012 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1013 					  0, MLX4_MCAST_DISABLE);
1014 		if (err)
1015 			en_err(priv, "Failed disabling multicast filter\n");
1016 
1017 		/* Add the default qp number as multicast promisc */
1018 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1019 			switch (mdev->dev->caps.steering_mode) {
1020 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1021 				err = mlx4_flow_steer_promisc_add(mdev->dev,
1022 								  priv->port,
1023 								  priv->base_qpn,
1024 								  MLX4_FS_MC_DEFAULT);
1025 				break;
1026 
1027 			case MLX4_STEERING_MODE_B0:
1028 				err = mlx4_multicast_promisc_add(mdev->dev,
1029 								 priv->base_qpn,
1030 								 priv->port);
1031 				break;
1032 
1033 			case MLX4_STEERING_MODE_A0:
1034 				break;
1035 			}
1036 			if (err)
1037 				en_err(priv, "Failed entering multicast promisc mode\n");
1038 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1039 		}
1040 	} else {
1041 		/* Disable Multicast promisc */
1042 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1043 			switch (mdev->dev->caps.steering_mode) {
1044 			case MLX4_STEERING_MODE_DEVICE_MANAGED:
1045 				err = mlx4_flow_steer_promisc_remove(mdev->dev,
1046 								     priv->port,
1047 								     MLX4_FS_MC_DEFAULT);
1048 				break;
1049 
1050 			case MLX4_STEERING_MODE_B0:
1051 				err = mlx4_multicast_promisc_remove(mdev->dev,
1052 								    priv->base_qpn,
1053 								    priv->port);
1054 				break;
1055 
1056 			case MLX4_STEERING_MODE_A0:
1057 				break;
1058 			}
1059 			if (err)
1060 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
1061 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1062 		}
1063 
1064 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1065 					  0, MLX4_MCAST_DISABLE);
1066 		if (err)
1067 			en_err(priv, "Failed disabling multicast filter\n");
1068 
1069 		/* Flush mcast filter and init it with broadcast address */
1070 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1071 				    1, MLX4_MCAST_CONFIG);
1072 
1073 		/* Update multicast list - we cache all addresses so they won't
1074 		 * change while HW is updated holding the command semaphor */
1075 		netif_addr_lock_bh(dev);
1076 		mlx4_en_cache_mclist(dev);
1077 		netif_addr_unlock_bh(dev);
1078 		list_for_each_entry(mclist, &priv->mc_list, list) {
1079 			mcast_addr = mlx4_mac_to_u64(mclist->addr);
1080 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1081 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
1082 		}
1083 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1084 					  0, MLX4_MCAST_ENABLE);
1085 		if (err)
1086 			en_err(priv, "Failed enabling multicast filter\n");
1087 
1088 		update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1089 		list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1090 			if (mclist->action == MCLIST_REM) {
1091 				/* detach this address and delete from list */
1092 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1093 				mc_list[5] = priv->port;
1094 				err = mlx4_multicast_detach(mdev->dev,
1095 							    priv->rss_map.indir_qp,
1096 							    mc_list,
1097 							    MLX4_PROT_ETH,
1098 							    mclist->reg_id);
1099 				if (err)
1100 					en_err(priv, "Fail to detach multicast address\n");
1101 
1102 				if (mclist->tunnel_reg_id) {
1103 					err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1104 					if (err)
1105 						en_err(priv, "Failed to detach multicast address\n");
1106 				}
1107 
1108 				/* remove from list */
1109 				list_del(&mclist->list);
1110 				kfree(mclist);
1111 			} else if (mclist->action == MCLIST_ADD) {
1112 				/* attach the address */
1113 				memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1114 				/* needed for B0 steering support */
1115 				mc_list[5] = priv->port;
1116 				err = mlx4_multicast_attach(mdev->dev,
1117 							    priv->rss_map.indir_qp,
1118 							    mc_list,
1119 							    priv->port, 0,
1120 							    MLX4_PROT_ETH,
1121 							    &mclist->reg_id);
1122 				if (err)
1123 					en_err(priv, "Fail to attach multicast address\n");
1124 
1125 				err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1126 							       &mclist->tunnel_reg_id);
1127 				if (err)
1128 					en_err(priv, "Failed to attach multicast address\n");
1129 			}
1130 		}
1131 	}
1132 }
1133 
mlx4_en_do_uc_filter(struct mlx4_en_priv * priv,struct net_device * dev,struct mlx4_en_dev * mdev)1134 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1135 				 struct net_device *dev,
1136 				 struct mlx4_en_dev *mdev)
1137 {
1138 	struct netdev_hw_addr *ha;
1139 	struct mlx4_mac_entry *entry;
1140 	struct hlist_node *tmp;
1141 	bool found;
1142 	u64 mac;
1143 	int err = 0;
1144 	struct hlist_head *bucket;
1145 	unsigned int i;
1146 	int removed = 0;
1147 	u32 prev_flags;
1148 
1149 	/* Note that we do not need to protect our mac_hash traversal with rcu,
1150 	 * since all modification code is protected by mdev->state_lock
1151 	 */
1152 
1153 	/* find what to remove */
1154 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1155 		bucket = &priv->mac_hash[i];
1156 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1157 			found = false;
1158 			netdev_for_each_uc_addr(ha, dev) {
1159 				if (ether_addr_equal_64bits(entry->mac,
1160 							    ha->addr)) {
1161 					found = true;
1162 					break;
1163 				}
1164 			}
1165 
1166 			/* MAC address of the port is not in uc list */
1167 			if (ether_addr_equal_64bits(entry->mac,
1168 						    priv->current_mac))
1169 				found = true;
1170 
1171 			if (!found) {
1172 				mac = mlx4_mac_to_u64(entry->mac);
1173 				mlx4_en_uc_steer_release(priv, entry->mac,
1174 							 priv->base_qpn,
1175 							 entry->reg_id);
1176 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1177 
1178 				hlist_del_rcu(&entry->hlist);
1179 				kfree_rcu(entry, rcu);
1180 				en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1181 				       entry->mac, priv->port);
1182 				++removed;
1183 			}
1184 		}
1185 	}
1186 
1187 	/* if we didn't remove anything, there is no use in trying to add
1188 	 * again once we are in a forced promisc mode state
1189 	 */
1190 	if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1191 		return;
1192 
1193 	prev_flags = priv->flags;
1194 	priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1195 
1196 	/* find what to add */
1197 	netdev_for_each_uc_addr(ha, dev) {
1198 		found = false;
1199 		bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1200 		hlist_for_each_entry(entry, bucket, hlist) {
1201 			if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1202 				found = true;
1203 				break;
1204 			}
1205 		}
1206 
1207 		if (!found) {
1208 			entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1209 			if (!entry) {
1210 				en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1211 				       ha->addr, priv->port);
1212 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1213 				break;
1214 			}
1215 			mac = mlx4_mac_to_u64(ha->addr);
1216 			memcpy(entry->mac, ha->addr, ETH_ALEN);
1217 			err = mlx4_register_mac(mdev->dev, priv->port, mac);
1218 			if (err < 0) {
1219 				en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1220 				       ha->addr, priv->port, err);
1221 				kfree(entry);
1222 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1223 				break;
1224 			}
1225 			err = mlx4_en_uc_steer_add(priv, ha->addr,
1226 						   &priv->base_qpn,
1227 						   &entry->reg_id);
1228 			if (err) {
1229 				en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1230 				       ha->addr, priv->port, err);
1231 				mlx4_unregister_mac(mdev->dev, priv->port, mac);
1232 				kfree(entry);
1233 				priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1234 				break;
1235 			} else {
1236 				unsigned int mac_hash;
1237 				en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1238 				       ha->addr, priv->port);
1239 				mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1240 				bucket = &priv->mac_hash[mac_hash];
1241 				hlist_add_head_rcu(&entry->hlist, bucket);
1242 			}
1243 		}
1244 	}
1245 
1246 	if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1247 		en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1248 			priv->port);
1249 	} else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1250 		en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1251 			priv->port);
1252 	}
1253 }
1254 
mlx4_en_do_set_rx_mode(struct work_struct * work)1255 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1256 {
1257 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1258 						 rx_mode_task);
1259 	struct mlx4_en_dev *mdev = priv->mdev;
1260 	struct net_device *dev = priv->dev;
1261 
1262 	mutex_lock(&mdev->state_lock);
1263 	if (!mdev->device_up) {
1264 		en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1265 		goto out;
1266 	}
1267 	if (!priv->port_up) {
1268 		en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1269 		goto out;
1270 	}
1271 
1272 	if (!netif_carrier_ok(dev)) {
1273 		if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1274 			if (priv->port_state.link_state) {
1275 				priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1276 				netif_carrier_on(dev);
1277 				en_dbg(LINK, priv, "Link Up\n");
1278 			}
1279 		}
1280 	}
1281 
1282 	if (dev->priv_flags & IFF_UNICAST_FLT)
1283 		mlx4_en_do_uc_filter(priv, dev, mdev);
1284 
1285 	/* Promsicuous mode: disable all filters */
1286 	if ((dev->flags & IFF_PROMISC) ||
1287 	    (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1288 		mlx4_en_set_promisc_mode(priv, mdev);
1289 		goto out;
1290 	}
1291 
1292 	/* Not in promiscuous mode */
1293 	if (priv->flags & MLX4_EN_FLAG_PROMISC)
1294 		mlx4_en_clear_promisc_mode(priv, mdev);
1295 
1296 	mlx4_en_do_multicast(priv, dev, mdev);
1297 out:
1298 	mutex_unlock(&mdev->state_lock);
1299 }
1300 
mlx4_en_set_rss_steer_rules(struct mlx4_en_priv * priv)1301 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1302 {
1303 	u64 reg_id;
1304 	int err = 0;
1305 	int *qpn = &priv->base_qpn;
1306 	struct mlx4_mac_entry *entry;
1307 
1308 	err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1309 	if (err)
1310 		return err;
1311 
1312 	err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1313 				       &priv->tunnel_reg_id);
1314 	if (err)
1315 		goto tunnel_err;
1316 
1317 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1318 	if (!entry) {
1319 		err = -ENOMEM;
1320 		goto alloc_err;
1321 	}
1322 
1323 	memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1324 	memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1325 	entry->reg_id = reg_id;
1326 	hlist_add_head_rcu(&entry->hlist,
1327 			   &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1328 
1329 	return 0;
1330 
1331 alloc_err:
1332 	if (priv->tunnel_reg_id)
1333 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1334 
1335 tunnel_err:
1336 	mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1337 	return err;
1338 }
1339 
mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv * priv)1340 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1341 {
1342 	u64 mac;
1343 	unsigned int i;
1344 	int qpn = priv->base_qpn;
1345 	struct hlist_head *bucket;
1346 	struct hlist_node *tmp;
1347 	struct mlx4_mac_entry *entry;
1348 
1349 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1350 		bucket = &priv->mac_hash[i];
1351 		hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1352 			mac = mlx4_mac_to_u64(entry->mac);
1353 			en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1354 			       entry->mac);
1355 			mlx4_en_uc_steer_release(priv, entry->mac,
1356 						 qpn, entry->reg_id);
1357 
1358 			mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1359 			hlist_del_rcu(&entry->hlist);
1360 			kfree_rcu(entry, rcu);
1361 		}
1362 	}
1363 
1364 	if (priv->tunnel_reg_id) {
1365 		mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1366 		priv->tunnel_reg_id = 0;
1367 	}
1368 }
1369 
mlx4_en_tx_timeout(struct net_device * dev,unsigned int txqueue)1370 static void mlx4_en_tx_timeout(struct net_device *dev, unsigned int txqueue)
1371 {
1372 	struct mlx4_en_priv *priv = netdev_priv(dev);
1373 	struct mlx4_en_dev *mdev = priv->mdev;
1374 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][txqueue];
1375 
1376 	if (netif_msg_timer(priv))
1377 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1378 
1379 	en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1380 		txqueue, tx_ring->qpn, tx_ring->sp_cqn,
1381 		tx_ring->cons, tx_ring->prod);
1382 
1383 	priv->port_stats.tx_timeout++;
1384 	if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state)) {
1385 		en_dbg(DRV, priv, "Scheduling port restart\n");
1386 		queue_work(mdev->workqueue, &priv->restart_task);
1387 	}
1388 }
1389 
1390 
1391 static void
mlx4_en_get_stats64(struct net_device * dev,struct rtnl_link_stats64 * stats)1392 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1393 {
1394 	struct mlx4_en_priv *priv = netdev_priv(dev);
1395 
1396 	spin_lock_bh(&priv->stats_lock);
1397 	mlx4_en_fold_software_stats(dev);
1398 	netdev_stats_to_stats64(stats, &dev->stats);
1399 	spin_unlock_bh(&priv->stats_lock);
1400 }
1401 
mlx4_en_set_default_moderation(struct mlx4_en_priv * priv)1402 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1403 {
1404 	struct mlx4_en_cq *cq;
1405 	int i, t;
1406 
1407 	/* If we haven't received a specific coalescing setting
1408 	 * (module param), we set the moderation parameters as follows:
1409 	 * - moder_cnt is set to the number of mtu sized packets to
1410 	 *   satisfy our coalescing target.
1411 	 * - moder_time is set to a fixed value.
1412 	 */
1413 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1414 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1415 	priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1416 	priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1417 	en_dbg(INTR, priv, "Default coalescing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1418 	       priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1419 
1420 	/* Setup cq moderation params */
1421 	for (i = 0; i < priv->rx_ring_num; i++) {
1422 		cq = priv->rx_cq[i];
1423 		cq->moder_cnt = priv->rx_frames;
1424 		cq->moder_time = priv->rx_usecs;
1425 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1426 		priv->last_moder_packets[i] = 0;
1427 		priv->last_moder_bytes[i] = 0;
1428 	}
1429 
1430 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1431 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1432 			cq = priv->tx_cq[t][i];
1433 			cq->moder_cnt = priv->tx_frames;
1434 			cq->moder_time = priv->tx_usecs;
1435 		}
1436 	}
1437 
1438 	/* Reset auto-moderation params */
1439 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1440 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1441 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1442 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1443 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1444 	priv->adaptive_rx_coal = 1;
1445 	priv->last_moder_jiffies = 0;
1446 	priv->last_moder_tx_packets = 0;
1447 }
1448 
mlx4_en_auto_moderation(struct mlx4_en_priv * priv)1449 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1450 {
1451 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1452 	u32 pkt_rate_high, pkt_rate_low;
1453 	struct mlx4_en_cq *cq;
1454 	unsigned long packets;
1455 	unsigned long rate;
1456 	unsigned long avg_pkt_size;
1457 	unsigned long rx_packets;
1458 	unsigned long rx_bytes;
1459 	unsigned long rx_pkt_diff;
1460 	int moder_time;
1461 	int ring, err;
1462 
1463 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1464 		return;
1465 
1466 	pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1467 	pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1468 
1469 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
1470 		rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1471 		rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1472 
1473 		rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1474 		packets = rx_pkt_diff;
1475 		rate = packets * HZ / period;
1476 		avg_pkt_size = packets ? (rx_bytes -
1477 				priv->last_moder_bytes[ring]) / packets : 0;
1478 
1479 		/* Apply auto-moderation only when packet rate
1480 		 * exceeds a rate that it matters */
1481 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1482 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1483 			if (rate <= pkt_rate_low)
1484 				moder_time = priv->rx_usecs_low;
1485 			else if (rate >= pkt_rate_high)
1486 				moder_time = priv->rx_usecs_high;
1487 			else
1488 				moder_time = (rate - pkt_rate_low) *
1489 					(priv->rx_usecs_high - priv->rx_usecs_low) /
1490 					(pkt_rate_high - pkt_rate_low) +
1491 					priv->rx_usecs_low;
1492 		} else {
1493 			moder_time = priv->rx_usecs_low;
1494 		}
1495 
1496 		cq = priv->rx_cq[ring];
1497 		if (moder_time != priv->last_moder_time[ring] ||
1498 		    cq->moder_cnt != priv->rx_frames) {
1499 			priv->last_moder_time[ring] = moder_time;
1500 			cq->moder_time = moder_time;
1501 			cq->moder_cnt = priv->rx_frames;
1502 			err = mlx4_en_set_cq_moder(priv, cq);
1503 			if (err)
1504 				en_err(priv, "Failed modifying moderation for cq:%d\n",
1505 				       ring);
1506 		}
1507 		priv->last_moder_packets[ring] = rx_packets;
1508 		priv->last_moder_bytes[ring] = rx_bytes;
1509 	}
1510 
1511 	priv->last_moder_jiffies = jiffies;
1512 }
1513 
mlx4_en_do_get_stats(struct work_struct * work)1514 static void mlx4_en_do_get_stats(struct work_struct *work)
1515 {
1516 	struct delayed_work *delay = to_delayed_work(work);
1517 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1518 						 stats_task);
1519 	struct mlx4_en_dev *mdev = priv->mdev;
1520 	int err;
1521 
1522 	mutex_lock(&mdev->state_lock);
1523 	if (mdev->device_up) {
1524 		if (priv->port_up) {
1525 			err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1526 			if (err)
1527 				en_dbg(HW, priv, "Could not update stats\n");
1528 
1529 			mlx4_en_auto_moderation(priv);
1530 		}
1531 
1532 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1533 	}
1534 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1535 		mlx4_en_do_set_mac(priv, priv->current_mac);
1536 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1537 	}
1538 	mutex_unlock(&mdev->state_lock);
1539 }
1540 
1541 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1542  * periodically
1543  */
mlx4_en_service_task(struct work_struct * work)1544 static void mlx4_en_service_task(struct work_struct *work)
1545 {
1546 	struct delayed_work *delay = to_delayed_work(work);
1547 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1548 						 service_task);
1549 	struct mlx4_en_dev *mdev = priv->mdev;
1550 
1551 	mutex_lock(&mdev->state_lock);
1552 	if (mdev->device_up) {
1553 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1554 			mlx4_en_ptp_overflow_check(mdev);
1555 
1556 		mlx4_en_recover_from_oom(priv);
1557 		queue_delayed_work(mdev->workqueue, &priv->service_task,
1558 				   SERVICE_TASK_DELAY);
1559 	}
1560 	mutex_unlock(&mdev->state_lock);
1561 }
1562 
mlx4_en_linkstate(struct work_struct * work)1563 static void mlx4_en_linkstate(struct work_struct *work)
1564 {
1565 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1566 						 linkstate_task);
1567 	struct mlx4_en_dev *mdev = priv->mdev;
1568 	int linkstate = priv->link_state;
1569 
1570 	mutex_lock(&mdev->state_lock);
1571 	/* If observable port state changed set carrier state and
1572 	 * report to system log */
1573 	if (priv->last_link_state != linkstate) {
1574 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1575 			en_info(priv, "Link Down\n");
1576 			netif_carrier_off(priv->dev);
1577 		} else {
1578 			en_info(priv, "Link Up\n");
1579 			netif_carrier_on(priv->dev);
1580 		}
1581 	}
1582 	priv->last_link_state = linkstate;
1583 	mutex_unlock(&mdev->state_lock);
1584 }
1585 
mlx4_en_init_affinity_hint(struct mlx4_en_priv * priv,int ring_idx)1586 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1587 {
1588 	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1589 	int numa_node = priv->mdev->dev->numa_node;
1590 
1591 	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1592 		return -ENOMEM;
1593 
1594 	cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1595 			ring->affinity_mask);
1596 	return 0;
1597 }
1598 
mlx4_en_free_affinity_hint(struct mlx4_en_priv * priv,int ring_idx)1599 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1600 {
1601 	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1602 }
1603 
mlx4_en_init_recycle_ring(struct mlx4_en_priv * priv,int tx_ring_idx)1604 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1605 				      int tx_ring_idx)
1606 {
1607 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1608 	int rr_index = tx_ring_idx;
1609 
1610 	tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1611 	tx_ring->recycle_ring = priv->rx_ring[rr_index];
1612 	en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1613 	       TX_XDP, tx_ring_idx, rr_index);
1614 }
1615 
mlx4_en_start_port(struct net_device * dev)1616 int mlx4_en_start_port(struct net_device *dev)
1617 {
1618 	struct mlx4_en_priv *priv = netdev_priv(dev);
1619 	struct mlx4_en_dev *mdev = priv->mdev;
1620 	struct mlx4_en_cq *cq;
1621 	struct mlx4_en_tx_ring *tx_ring;
1622 	int rx_index = 0;
1623 	int err = 0;
1624 	int i, t;
1625 	int j;
1626 	u8 mc_list[16] = {0};
1627 
1628 	if (priv->port_up) {
1629 		en_dbg(DRV, priv, "start port called while port already up\n");
1630 		return 0;
1631 	}
1632 
1633 	INIT_LIST_HEAD(&priv->mc_list);
1634 	INIT_LIST_HEAD(&priv->curr_list);
1635 	INIT_LIST_HEAD(&priv->ethtool_list);
1636 	memset(&priv->ethtool_rules[0], 0,
1637 	       sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1638 
1639 	/* Calculate Rx buf size */
1640 	dev->mtu = min(dev->mtu, priv->max_mtu);
1641 	mlx4_en_calc_rx_buf(dev);
1642 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1643 
1644 	/* Configure rx cq's and rings */
1645 	err = mlx4_en_activate_rx_rings(priv);
1646 	if (err) {
1647 		en_err(priv, "Failed to activate RX rings\n");
1648 		return err;
1649 	}
1650 	for (i = 0; i < priv->rx_ring_num; i++) {
1651 		cq = priv->rx_cq[i];
1652 
1653 		err = mlx4_en_init_affinity_hint(priv, i);
1654 		if (err) {
1655 			en_err(priv, "Failed preparing IRQ affinity hint\n");
1656 			goto cq_err;
1657 		}
1658 
1659 		err = mlx4_en_activate_cq(priv, cq, i);
1660 		if (err) {
1661 			en_err(priv, "Failed activating Rx CQ\n");
1662 			mlx4_en_free_affinity_hint(priv, i);
1663 			goto cq_err;
1664 		}
1665 
1666 		for (j = 0; j < cq->size; j++) {
1667 			struct mlx4_cqe *cqe = NULL;
1668 
1669 			cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1670 			      priv->cqe_factor;
1671 			cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1672 		}
1673 
1674 		err = mlx4_en_set_cq_moder(priv, cq);
1675 		if (err) {
1676 			en_err(priv, "Failed setting cq moderation parameters\n");
1677 			mlx4_en_deactivate_cq(priv, cq);
1678 			mlx4_en_free_affinity_hint(priv, i);
1679 			goto cq_err;
1680 		}
1681 		mlx4_en_arm_cq(priv, cq);
1682 		priv->rx_ring[i]->cqn = cq->mcq.cqn;
1683 		++rx_index;
1684 	}
1685 
1686 	/* Set qp number */
1687 	en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1688 	err = mlx4_en_get_qp(priv);
1689 	if (err) {
1690 		en_err(priv, "Failed getting eth qp\n");
1691 		goto cq_err;
1692 	}
1693 	mdev->mac_removed[priv->port] = 0;
1694 
1695 	priv->counter_index =
1696 			mlx4_get_default_counter_index(mdev->dev, priv->port);
1697 
1698 	err = mlx4_en_config_rss_steer(priv);
1699 	if (err) {
1700 		en_err(priv, "Failed configuring rss steering\n");
1701 		goto mac_err;
1702 	}
1703 
1704 	err = mlx4_en_create_drop_qp(priv);
1705 	if (err)
1706 		goto rss_err;
1707 
1708 	/* Configure tx cq's and rings */
1709 	for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1710 		u8 num_tx_rings_p_up = t == TX ?
1711 			priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1712 
1713 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1714 			/* Configure cq */
1715 			cq = priv->tx_cq[t][i];
1716 			err = mlx4_en_activate_cq(priv, cq, i);
1717 			if (err) {
1718 				en_err(priv, "Failed allocating Tx CQ\n");
1719 				goto tx_err;
1720 			}
1721 			err = mlx4_en_set_cq_moder(priv, cq);
1722 			if (err) {
1723 				en_err(priv, "Failed setting cq moderation parameters\n");
1724 				mlx4_en_deactivate_cq(priv, cq);
1725 				goto tx_err;
1726 			}
1727 			en_dbg(DRV, priv,
1728 			       "Resetting index of collapsed CQ:%d to -1\n", i);
1729 			cq->buf->wqe_index = cpu_to_be16(0xffff);
1730 
1731 			/* Configure ring */
1732 			tx_ring = priv->tx_ring[t][i];
1733 			err = mlx4_en_activate_tx_ring(priv, tx_ring,
1734 						       cq->mcq.cqn,
1735 						       i / num_tx_rings_p_up);
1736 			if (err) {
1737 				en_err(priv, "Failed allocating Tx ring\n");
1738 				mlx4_en_deactivate_cq(priv, cq);
1739 				goto tx_err;
1740 			}
1741 			clear_bit(MLX4_EN_TX_RING_STATE_RECOVERING, &tx_ring->state);
1742 			if (t != TX_XDP) {
1743 				tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1744 				tx_ring->recycle_ring = NULL;
1745 
1746 				/* Arm CQ for TX completions */
1747 				mlx4_en_arm_cq(priv, cq);
1748 
1749 			} else {
1750 				mlx4_en_init_tx_xdp_ring_descs(priv, tx_ring);
1751 				mlx4_en_init_recycle_ring(priv, i);
1752 				/* XDP TX CQ should never be armed */
1753 			}
1754 
1755 			/* Set initial ownership of all Tx TXBBs to SW (1) */
1756 			for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1757 				*((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1758 		}
1759 	}
1760 
1761 	/* Configure port */
1762 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1763 				    priv->rx_skb_size + ETH_FCS_LEN,
1764 				    priv->prof->tx_pause,
1765 				    priv->prof->tx_ppp,
1766 				    priv->prof->rx_pause,
1767 				    priv->prof->rx_ppp);
1768 	if (err) {
1769 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1770 		       priv->port, err);
1771 		goto tx_err;
1772 	}
1773 
1774 	err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1775 	if (err) {
1776 		en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1777 		       dev->mtu, priv->port, err);
1778 		goto tx_err;
1779 	}
1780 
1781 	/* Set default qp number */
1782 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1783 	if (err) {
1784 		en_err(priv, "Failed setting default qp numbers\n");
1785 		goto tx_err;
1786 	}
1787 
1788 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1789 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1790 		if (err) {
1791 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1792 			       err);
1793 			goto tx_err;
1794 		}
1795 	}
1796 
1797 	/* Init port */
1798 	en_dbg(HW, priv, "Initializing port\n");
1799 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1800 	if (err) {
1801 		en_err(priv, "Failed Initializing port\n");
1802 		goto tx_err;
1803 	}
1804 
1805 	/* Set Unicast and VXLAN steering rules */
1806 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1807 	    mlx4_en_set_rss_steer_rules(priv))
1808 		mlx4_warn(mdev, "Failed setting steering rules\n");
1809 
1810 	/* Attach rx QP to bradcast address */
1811 	eth_broadcast_addr(&mc_list[10]);
1812 	mc_list[5] = priv->port; /* needed for B0 steering support */
1813 	if (mlx4_multicast_attach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1814 				  priv->port, 0, MLX4_PROT_ETH,
1815 				  &priv->broadcast_id))
1816 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1817 
1818 	/* Must redo promiscuous mode setup. */
1819 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1820 
1821 	/* Schedule multicast task to populate multicast list */
1822 	queue_work(mdev->workqueue, &priv->rx_mode_task);
1823 
1824 	if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1825 		udp_tunnel_nic_reset_ntf(dev);
1826 
1827 	priv->port_up = true;
1828 
1829 	/* Process all completions if exist to prevent
1830 	 * the queues freezing if they are full
1831 	 */
1832 	for (i = 0; i < priv->rx_ring_num; i++) {
1833 		local_bh_disable();
1834 		napi_schedule(&priv->rx_cq[i]->napi);
1835 		local_bh_enable();
1836 	}
1837 
1838 	clear_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state);
1839 	netif_tx_start_all_queues(dev);
1840 	netif_device_attach(dev);
1841 
1842 	return 0;
1843 
1844 tx_err:
1845 	if (t == MLX4_EN_NUM_TX_TYPES) {
1846 		t--;
1847 		i = priv->tx_ring_num[t];
1848 	}
1849 	while (t >= 0) {
1850 		while (i--) {
1851 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1852 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1853 		}
1854 		if (!t--)
1855 			break;
1856 		i = priv->tx_ring_num[t];
1857 	}
1858 	mlx4_en_destroy_drop_qp(priv);
1859 rss_err:
1860 	mlx4_en_release_rss_steer(priv);
1861 mac_err:
1862 	mlx4_en_put_qp(priv);
1863 cq_err:
1864 	while (rx_index--) {
1865 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1866 		mlx4_en_free_affinity_hint(priv, rx_index);
1867 	}
1868 	for (i = 0; i < priv->rx_ring_num; i++)
1869 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1870 
1871 	return err; /* need to close devices */
1872 }
1873 
1874 
mlx4_en_stop_port(struct net_device * dev,int detach)1875 void mlx4_en_stop_port(struct net_device *dev, int detach)
1876 {
1877 	struct mlx4_en_priv *priv = netdev_priv(dev);
1878 	struct mlx4_en_dev *mdev = priv->mdev;
1879 	struct mlx4_en_mc_list *mclist, *tmp;
1880 	struct ethtool_flow_id *flow, *tmp_flow;
1881 	int i, t;
1882 	u8 mc_list[16] = {0};
1883 
1884 	if (!priv->port_up) {
1885 		en_dbg(DRV, priv, "stop port called while port already down\n");
1886 		return;
1887 	}
1888 
1889 	/* close port*/
1890 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
1891 
1892 	/* Synchronize with tx routine */
1893 	netif_tx_lock_bh(dev);
1894 	if (detach)
1895 		netif_device_detach(dev);
1896 	netif_tx_stop_all_queues(dev);
1897 	netif_tx_unlock_bh(dev);
1898 
1899 	netif_tx_disable(dev);
1900 
1901 	spin_lock_bh(&priv->stats_lock);
1902 	mlx4_en_fold_software_stats(dev);
1903 	/* Set port as not active */
1904 	priv->port_up = false;
1905 	spin_unlock_bh(&priv->stats_lock);
1906 
1907 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1908 
1909 	/* Promsicuous mode */
1910 	if (mdev->dev->caps.steering_mode ==
1911 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1912 		priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1913 				 MLX4_EN_FLAG_MC_PROMISC);
1914 		mlx4_flow_steer_promisc_remove(mdev->dev,
1915 					       priv->port,
1916 					       MLX4_FS_ALL_DEFAULT);
1917 		mlx4_flow_steer_promisc_remove(mdev->dev,
1918 					       priv->port,
1919 					       MLX4_FS_MC_DEFAULT);
1920 	} else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1921 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1922 
1923 		/* Disable promiscouos mode */
1924 		mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1925 					    priv->port);
1926 
1927 		/* Disable Multicast promisc */
1928 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1929 			mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1930 						      priv->port);
1931 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1932 		}
1933 	}
1934 
1935 	/* Detach All multicasts */
1936 	eth_broadcast_addr(&mc_list[10]);
1937 	mc_list[5] = priv->port; /* needed for B0 steering support */
1938 	mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1939 			      MLX4_PROT_ETH, priv->broadcast_id);
1940 	list_for_each_entry(mclist, &priv->curr_list, list) {
1941 		memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1942 		mc_list[5] = priv->port;
1943 		mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp,
1944 				      mc_list, MLX4_PROT_ETH, mclist->reg_id);
1945 		if (mclist->tunnel_reg_id)
1946 			mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1947 	}
1948 	mlx4_en_clear_list(dev);
1949 	list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1950 		list_del(&mclist->list);
1951 		kfree(mclist);
1952 	}
1953 
1954 	/* Flush multicast filter */
1955 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1956 
1957 	/* Remove flow steering rules for the port*/
1958 	if (mdev->dev->caps.steering_mode ==
1959 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1960 		ASSERT_RTNL();
1961 		list_for_each_entry_safe(flow, tmp_flow,
1962 					 &priv->ethtool_list, list) {
1963 			mlx4_flow_detach(mdev->dev, flow->id);
1964 			list_del(&flow->list);
1965 		}
1966 	}
1967 
1968 	mlx4_en_destroy_drop_qp(priv);
1969 
1970 	/* Free TX Rings */
1971 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1972 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
1973 			mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1974 			mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1975 		}
1976 	}
1977 	msleep(10);
1978 
1979 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1980 		for (i = 0; i < priv->tx_ring_num[t]; i++)
1981 			mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1982 
1983 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1984 		mlx4_en_delete_rss_steer_rules(priv);
1985 
1986 	/* Free RSS qps */
1987 	mlx4_en_release_rss_steer(priv);
1988 
1989 	/* Unregister Mac address for the port */
1990 	mlx4_en_put_qp(priv);
1991 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1992 		mdev->mac_removed[priv->port] = 1;
1993 
1994 	/* Free RX Rings */
1995 	for (i = 0; i < priv->rx_ring_num; i++) {
1996 		struct mlx4_en_cq *cq = priv->rx_cq[i];
1997 
1998 		napi_synchronize(&cq->napi);
1999 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
2000 		mlx4_en_deactivate_cq(priv, cq);
2001 
2002 		mlx4_en_free_affinity_hint(priv, i);
2003 	}
2004 }
2005 
mlx4_en_restart(struct work_struct * work)2006 static void mlx4_en_restart(struct work_struct *work)
2007 {
2008 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2009 						 restart_task);
2010 	struct mlx4_en_dev *mdev = priv->mdev;
2011 	struct net_device *dev = priv->dev;
2012 
2013 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
2014 
2015 	rtnl_lock();
2016 	mutex_lock(&mdev->state_lock);
2017 	if (priv->port_up) {
2018 		mlx4_en_stop_port(dev, 1);
2019 		if (mlx4_en_start_port(dev))
2020 			en_err(priv, "Failed restarting port %d\n", priv->port);
2021 	}
2022 	mutex_unlock(&mdev->state_lock);
2023 	rtnl_unlock();
2024 }
2025 
mlx4_en_clear_stats(struct net_device * dev)2026 static void mlx4_en_clear_stats(struct net_device *dev)
2027 {
2028 	struct mlx4_en_priv *priv = netdev_priv(dev);
2029 	struct mlx4_en_dev *mdev = priv->mdev;
2030 	struct mlx4_en_tx_ring **tx_ring;
2031 	int i;
2032 
2033 	if (!mlx4_is_slave(mdev->dev))
2034 		if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
2035 			en_dbg(HW, priv, "Failed dumping statistics\n");
2036 
2037 	memset(&priv->pstats, 0, sizeof(priv->pstats));
2038 	memset(&priv->pkstats, 0, sizeof(priv->pkstats));
2039 	memset(&priv->port_stats, 0, sizeof(priv->port_stats));
2040 	memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
2041 	memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
2042 	memset(&priv->rx_priority_flowstats, 0,
2043 	       sizeof(priv->rx_priority_flowstats));
2044 	memset(&priv->tx_priority_flowstats, 0,
2045 	       sizeof(priv->tx_priority_flowstats));
2046 	memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
2047 
2048 	tx_ring = priv->tx_ring[TX];
2049 	for (i = 0; i < priv->tx_ring_num[TX]; i++) {
2050 		tx_ring[i]->bytes = 0;
2051 		tx_ring[i]->packets = 0;
2052 		tx_ring[i]->tx_csum = 0;
2053 		tx_ring[i]->tx_dropped = 0;
2054 		tx_ring[i]->queue_stopped = 0;
2055 		tx_ring[i]->wake_queue = 0;
2056 		tx_ring[i]->tso_packets = 0;
2057 		tx_ring[i]->xmit_more = 0;
2058 	}
2059 	for (i = 0; i < priv->rx_ring_num; i++) {
2060 		priv->rx_ring[i]->bytes = 0;
2061 		priv->rx_ring[i]->packets = 0;
2062 		priv->rx_ring[i]->csum_ok = 0;
2063 		priv->rx_ring[i]->csum_none = 0;
2064 		priv->rx_ring[i]->csum_complete = 0;
2065 	}
2066 }
2067 
mlx4_en_open(struct net_device * dev)2068 static int mlx4_en_open(struct net_device *dev)
2069 {
2070 	struct mlx4_en_priv *priv = netdev_priv(dev);
2071 	struct mlx4_en_dev *mdev = priv->mdev;
2072 	int err = 0;
2073 
2074 	mutex_lock(&mdev->state_lock);
2075 
2076 	if (!mdev->device_up) {
2077 		en_err(priv, "Cannot open - device down/disabled\n");
2078 		err = -EBUSY;
2079 		goto out;
2080 	}
2081 
2082 	/* Reset HW statistics and SW counters */
2083 	mlx4_en_clear_stats(dev);
2084 
2085 	err = mlx4_en_start_port(dev);
2086 	if (err)
2087 		en_err(priv, "Failed starting port:%d\n", priv->port);
2088 
2089 out:
2090 	mutex_unlock(&mdev->state_lock);
2091 	return err;
2092 }
2093 
2094 
mlx4_en_close(struct net_device * dev)2095 static int mlx4_en_close(struct net_device *dev)
2096 {
2097 	struct mlx4_en_priv *priv = netdev_priv(dev);
2098 	struct mlx4_en_dev *mdev = priv->mdev;
2099 
2100 	en_dbg(IFDOWN, priv, "Close port called\n");
2101 
2102 	mutex_lock(&mdev->state_lock);
2103 
2104 	mlx4_en_stop_port(dev, 0);
2105 	netif_carrier_off(dev);
2106 
2107 	mutex_unlock(&mdev->state_lock);
2108 	return 0;
2109 }
2110 
mlx4_en_free_resources(struct mlx4_en_priv * priv)2111 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2112 {
2113 	int i, t;
2114 
2115 #ifdef CONFIG_RFS_ACCEL
2116 	priv->dev->rx_cpu_rmap = NULL;
2117 #endif
2118 
2119 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2120 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2121 			if (priv->tx_ring[t] && priv->tx_ring[t][i])
2122 				mlx4_en_destroy_tx_ring(priv,
2123 							&priv->tx_ring[t][i]);
2124 			if (priv->tx_cq[t] && priv->tx_cq[t][i])
2125 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2126 		}
2127 		kfree(priv->tx_ring[t]);
2128 		kfree(priv->tx_cq[t]);
2129 	}
2130 
2131 	for (i = 0; i < priv->rx_ring_num; i++) {
2132 		if (priv->rx_ring[i])
2133 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2134 				priv->prof->rx_ring_size, priv->stride);
2135 		if (priv->rx_cq[i])
2136 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2137 	}
2138 
2139 }
2140 
mlx4_en_alloc_resources(struct mlx4_en_priv * priv)2141 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2142 {
2143 	struct mlx4_en_port_profile *prof = priv->prof;
2144 	int i, t;
2145 	int node;
2146 
2147 	/* Create tx Rings */
2148 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2149 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2150 			node = cpu_to_node(i % num_online_cpus());
2151 			if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2152 					      prof->tx_ring_size, i, t, node))
2153 				goto err;
2154 
2155 			if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2156 						   prof->tx_ring_size,
2157 						   TXBB_SIZE, node, i))
2158 				goto err;
2159 		}
2160 	}
2161 
2162 	/* Create rx Rings */
2163 	for (i = 0; i < priv->rx_ring_num; i++) {
2164 		node = cpu_to_node(i % num_online_cpus());
2165 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2166 				      prof->rx_ring_size, i, RX, node))
2167 			goto err;
2168 
2169 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2170 					   prof->rx_ring_size, priv->stride,
2171 					   node, i))
2172 			goto err;
2173 
2174 	}
2175 
2176 #ifdef CONFIG_RFS_ACCEL
2177 	priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2178 #endif
2179 
2180 	return 0;
2181 
2182 err:
2183 	en_err(priv, "Failed to allocate NIC resources\n");
2184 	for (i = 0; i < priv->rx_ring_num; i++) {
2185 		if (priv->rx_ring[i])
2186 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2187 						prof->rx_ring_size,
2188 						priv->stride);
2189 		if (priv->rx_cq[i])
2190 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2191 	}
2192 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2193 		for (i = 0; i < priv->tx_ring_num[t]; i++) {
2194 			if (priv->tx_ring[t][i])
2195 				mlx4_en_destroy_tx_ring(priv,
2196 							&priv->tx_ring[t][i]);
2197 			if (priv->tx_cq[t][i])
2198 				mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2199 		}
2200 	}
2201 	return -ENOMEM;
2202 }
2203 
2204 
mlx4_en_copy_priv(struct mlx4_en_priv * dst,struct mlx4_en_priv * src,struct mlx4_en_port_profile * prof)2205 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2206 			     struct mlx4_en_priv *src,
2207 			     struct mlx4_en_port_profile *prof)
2208 {
2209 	int t;
2210 
2211 	memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2212 	       sizeof(dst->hwtstamp_config));
2213 	dst->num_tx_rings_p_up = prof->num_tx_rings_p_up;
2214 	dst->rx_ring_num = prof->rx_ring_num;
2215 	dst->flags = prof->flags;
2216 	dst->mdev = src->mdev;
2217 	dst->port = src->port;
2218 	dst->dev = src->dev;
2219 	dst->prof = prof;
2220 	dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2221 					 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2222 
2223 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2224 		dst->tx_ring_num[t] = prof->tx_ring_num[t];
2225 		if (!dst->tx_ring_num[t])
2226 			continue;
2227 
2228 		dst->tx_ring[t] = kcalloc(MAX_TX_RINGS,
2229 					  sizeof(struct mlx4_en_tx_ring *),
2230 					  GFP_KERNEL);
2231 		if (!dst->tx_ring[t])
2232 			goto err_free_tx;
2233 
2234 		dst->tx_cq[t] = kcalloc(MAX_TX_RINGS,
2235 					sizeof(struct mlx4_en_cq *),
2236 					GFP_KERNEL);
2237 		if (!dst->tx_cq[t]) {
2238 			kfree(dst->tx_ring[t]);
2239 			goto err_free_tx;
2240 		}
2241 	}
2242 
2243 	return 0;
2244 
2245 err_free_tx:
2246 	while (t--) {
2247 		kfree(dst->tx_ring[t]);
2248 		kfree(dst->tx_cq[t]);
2249 	}
2250 	return -ENOMEM;
2251 }
2252 
mlx4_en_update_priv(struct mlx4_en_priv * dst,struct mlx4_en_priv * src)2253 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2254 				struct mlx4_en_priv *src)
2255 {
2256 	int t;
2257 	memcpy(dst->rx_ring, src->rx_ring,
2258 	       sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2259 	memcpy(dst->rx_cq, src->rx_cq,
2260 	       sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2261 	memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2262 	       sizeof(dst->hwtstamp_config));
2263 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2264 		dst->tx_ring_num[t] = src->tx_ring_num[t];
2265 		dst->tx_ring[t] = src->tx_ring[t];
2266 		dst->tx_cq[t] = src->tx_cq[t];
2267 	}
2268 	dst->num_tx_rings_p_up = src->num_tx_rings_p_up;
2269 	dst->rx_ring_num = src->rx_ring_num;
2270 	memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2271 }
2272 
mlx4_en_try_alloc_resources(struct mlx4_en_priv * priv,struct mlx4_en_priv * tmp,struct mlx4_en_port_profile * prof,bool carry_xdp_prog)2273 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2274 				struct mlx4_en_priv *tmp,
2275 				struct mlx4_en_port_profile *prof,
2276 				bool carry_xdp_prog)
2277 {
2278 	struct bpf_prog *xdp_prog;
2279 	int i, t, ret;
2280 
2281 	ret = mlx4_en_copy_priv(tmp, priv, prof);
2282 	if (ret) {
2283 		en_warn(priv, "%s: mlx4_en_copy_priv() failed, return\n",
2284 			__func__);
2285 		return ret;
2286 	}
2287 
2288 	if (mlx4_en_alloc_resources(tmp)) {
2289 		en_warn(priv,
2290 			"%s: Resource allocation failed, using previous configuration\n",
2291 			__func__);
2292 		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2293 			kfree(tmp->tx_ring[t]);
2294 			kfree(tmp->tx_cq[t]);
2295 		}
2296 		return -ENOMEM;
2297 	}
2298 
2299 	/* All rx_rings has the same xdp_prog.  Pick the first one. */
2300 	xdp_prog = rcu_dereference_protected(
2301 		priv->rx_ring[0]->xdp_prog,
2302 		lockdep_is_held(&priv->mdev->state_lock));
2303 
2304 	if (xdp_prog && carry_xdp_prog) {
2305 		bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2306 		for (i = 0; i < tmp->rx_ring_num; i++)
2307 			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2308 					   xdp_prog);
2309 	}
2310 
2311 	return 0;
2312 }
2313 
mlx4_en_safe_replace_resources(struct mlx4_en_priv * priv,struct mlx4_en_priv * tmp)2314 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2315 				    struct mlx4_en_priv *tmp)
2316 {
2317 	mlx4_en_free_resources(priv);
2318 	mlx4_en_update_priv(priv, tmp);
2319 }
2320 
mlx4_en_destroy_netdev(struct net_device * dev)2321 void mlx4_en_destroy_netdev(struct net_device *dev)
2322 {
2323 	struct mlx4_en_priv *priv = netdev_priv(dev);
2324 	struct mlx4_en_dev *mdev = priv->mdev;
2325 
2326 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2327 
2328 	/* Unregister device - this will close the port if it was up */
2329 	if (priv->registered) {
2330 		devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2331 							      priv->port));
2332 		unregister_netdev(dev);
2333 	}
2334 
2335 	if (priv->allocated)
2336 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2337 
2338 	cancel_delayed_work(&priv->stats_task);
2339 	cancel_delayed_work(&priv->service_task);
2340 	/* flush any pending task for this netdev */
2341 	flush_workqueue(mdev->workqueue);
2342 
2343 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2344 		mlx4_en_remove_timestamp(mdev);
2345 
2346 	/* Detach the netdev so tasks would not attempt to access it */
2347 	mutex_lock(&mdev->state_lock);
2348 	mdev->pndev[priv->port] = NULL;
2349 	mdev->upper[priv->port] = NULL;
2350 
2351 #ifdef CONFIG_RFS_ACCEL
2352 	mlx4_en_cleanup_filters(priv);
2353 #endif
2354 
2355 	mlx4_en_free_resources(priv);
2356 	mutex_unlock(&mdev->state_lock);
2357 
2358 	free_netdev(dev);
2359 }
2360 
mlx4_en_check_xdp_mtu(struct net_device * dev,int mtu)2361 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2362 {
2363 	struct mlx4_en_priv *priv = netdev_priv(dev);
2364 
2365 	if (mtu > MLX4_EN_MAX_XDP_MTU) {
2366 		en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2367 		       mtu, MLX4_EN_MAX_XDP_MTU);
2368 		return false;
2369 	}
2370 
2371 	return true;
2372 }
2373 
mlx4_en_change_mtu(struct net_device * dev,int new_mtu)2374 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2375 {
2376 	struct mlx4_en_priv *priv = netdev_priv(dev);
2377 	struct mlx4_en_dev *mdev = priv->mdev;
2378 	int err = 0;
2379 
2380 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2381 		 dev->mtu, new_mtu);
2382 
2383 	if (priv->tx_ring_num[TX_XDP] &&
2384 	    !mlx4_en_check_xdp_mtu(dev, new_mtu))
2385 		return -EOPNOTSUPP;
2386 
2387 	dev->mtu = new_mtu;
2388 
2389 	if (netif_running(dev)) {
2390 		mutex_lock(&mdev->state_lock);
2391 		if (!mdev->device_up) {
2392 			/* NIC is probably restarting - let restart task reset
2393 			 * the port */
2394 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2395 		} else {
2396 			mlx4_en_stop_port(dev, 1);
2397 			err = mlx4_en_start_port(dev);
2398 			if (err) {
2399 				en_err(priv, "Failed restarting port:%d\n",
2400 					 priv->port);
2401 				if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING,
2402 						      &priv->state))
2403 					queue_work(mdev->workqueue, &priv->restart_task);
2404 			}
2405 		}
2406 		mutex_unlock(&mdev->state_lock);
2407 	}
2408 	return 0;
2409 }
2410 
mlx4_en_hwtstamp_set(struct net_device * dev,struct ifreq * ifr)2411 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2412 {
2413 	struct mlx4_en_priv *priv = netdev_priv(dev);
2414 	struct mlx4_en_dev *mdev = priv->mdev;
2415 	struct hwtstamp_config config;
2416 
2417 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2418 		return -EFAULT;
2419 
2420 	/* reserved for future extensions */
2421 	if (config.flags)
2422 		return -EINVAL;
2423 
2424 	/* device doesn't support time stamping */
2425 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2426 		return -EINVAL;
2427 
2428 	/* TX HW timestamp */
2429 	switch (config.tx_type) {
2430 	case HWTSTAMP_TX_OFF:
2431 	case HWTSTAMP_TX_ON:
2432 		break;
2433 	default:
2434 		return -ERANGE;
2435 	}
2436 
2437 	/* RX HW timestamp */
2438 	switch (config.rx_filter) {
2439 	case HWTSTAMP_FILTER_NONE:
2440 		break;
2441 	case HWTSTAMP_FILTER_ALL:
2442 	case HWTSTAMP_FILTER_SOME:
2443 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2444 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2445 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2446 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2447 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2448 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2449 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2450 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2451 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2452 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
2453 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
2454 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2455 	case HWTSTAMP_FILTER_NTP_ALL:
2456 		config.rx_filter = HWTSTAMP_FILTER_ALL;
2457 		break;
2458 	default:
2459 		return -ERANGE;
2460 	}
2461 
2462 	if (mlx4_en_reset_config(dev, config, dev->features)) {
2463 		config.tx_type = HWTSTAMP_TX_OFF;
2464 		config.rx_filter = HWTSTAMP_FILTER_NONE;
2465 	}
2466 
2467 	return copy_to_user(ifr->ifr_data, &config,
2468 			    sizeof(config)) ? -EFAULT : 0;
2469 }
2470 
mlx4_en_hwtstamp_get(struct net_device * dev,struct ifreq * ifr)2471 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2472 {
2473 	struct mlx4_en_priv *priv = netdev_priv(dev);
2474 
2475 	return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2476 			    sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2477 }
2478 
mlx4_en_ioctl(struct net_device * dev,struct ifreq * ifr,int cmd)2479 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2480 {
2481 	switch (cmd) {
2482 	case SIOCSHWTSTAMP:
2483 		return mlx4_en_hwtstamp_set(dev, ifr);
2484 	case SIOCGHWTSTAMP:
2485 		return mlx4_en_hwtstamp_get(dev, ifr);
2486 	default:
2487 		return -EOPNOTSUPP;
2488 	}
2489 }
2490 
mlx4_en_fix_features(struct net_device * netdev,netdev_features_t features)2491 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2492 					      netdev_features_t features)
2493 {
2494 	struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2495 	struct mlx4_en_dev *mdev = en_priv->mdev;
2496 
2497 	/* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2498 	 * enable/disable make sure S-TAG flag is always in same state as
2499 	 * C-TAG.
2500 	 */
2501 	if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2502 	    !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2503 		features |= NETIF_F_HW_VLAN_STAG_RX;
2504 	else
2505 		features &= ~NETIF_F_HW_VLAN_STAG_RX;
2506 
2507 	return features;
2508 }
2509 
mlx4_en_set_features(struct net_device * netdev,netdev_features_t features)2510 static int mlx4_en_set_features(struct net_device *netdev,
2511 		netdev_features_t features)
2512 {
2513 	struct mlx4_en_priv *priv = netdev_priv(netdev);
2514 	bool reset = false;
2515 	int ret = 0;
2516 
2517 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2518 		en_info(priv, "Turn %s RX-FCS\n",
2519 			(features & NETIF_F_RXFCS) ? "ON" : "OFF");
2520 		reset = true;
2521 	}
2522 
2523 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2524 		u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2525 
2526 		en_info(priv, "Turn %s RX-ALL\n",
2527 			ignore_fcs_value ? "ON" : "OFF");
2528 		ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2529 					      priv->port, ignore_fcs_value);
2530 		if (ret)
2531 			return ret;
2532 	}
2533 
2534 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2535 		en_info(priv, "Turn %s RX vlan strip offload\n",
2536 			(features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2537 		reset = true;
2538 	}
2539 
2540 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2541 		en_info(priv, "Turn %s TX vlan strip offload\n",
2542 			(features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2543 
2544 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2545 		en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2546 			(features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2547 
2548 	if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2549 		en_info(priv, "Turn %s loopback\n",
2550 			(features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2551 		mlx4_en_update_loopback_state(netdev, features);
2552 	}
2553 
2554 	if (reset) {
2555 		ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2556 					   features);
2557 		if (ret)
2558 			return ret;
2559 	}
2560 
2561 	return 0;
2562 }
2563 
mlx4_en_set_vf_mac(struct net_device * dev,int queue,u8 * mac)2564 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2565 {
2566 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2567 	struct mlx4_en_dev *mdev = en_priv->mdev;
2568 
2569 	return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac);
2570 }
2571 
mlx4_en_set_vf_vlan(struct net_device * dev,int vf,u16 vlan,u8 qos,__be16 vlan_proto)2572 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2573 			       __be16 vlan_proto)
2574 {
2575 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2576 	struct mlx4_en_dev *mdev = en_priv->mdev;
2577 
2578 	return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2579 				vlan_proto);
2580 }
2581 
mlx4_en_set_vf_rate(struct net_device * dev,int vf,int min_tx_rate,int max_tx_rate)2582 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2583 			       int max_tx_rate)
2584 {
2585 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2586 	struct mlx4_en_dev *mdev = en_priv->mdev;
2587 
2588 	return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2589 				max_tx_rate);
2590 }
2591 
mlx4_en_set_vf_spoofchk(struct net_device * dev,int vf,bool setting)2592 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2593 {
2594 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2595 	struct mlx4_en_dev *mdev = en_priv->mdev;
2596 
2597 	return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2598 }
2599 
mlx4_en_get_vf_config(struct net_device * dev,int vf,struct ifla_vf_info * ivf)2600 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2601 {
2602 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2603 	struct mlx4_en_dev *mdev = en_priv->mdev;
2604 
2605 	return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2606 }
2607 
mlx4_en_set_vf_link_state(struct net_device * dev,int vf,int link_state)2608 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2609 {
2610 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2611 	struct mlx4_en_dev *mdev = en_priv->mdev;
2612 
2613 	return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2614 }
2615 
mlx4_en_get_vf_stats(struct net_device * dev,int vf,struct ifla_vf_stats * vf_stats)2616 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2617 				struct ifla_vf_stats *vf_stats)
2618 {
2619 	struct mlx4_en_priv *en_priv = netdev_priv(dev);
2620 	struct mlx4_en_dev *mdev = en_priv->mdev;
2621 
2622 	return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2623 }
2624 
2625 #define PORT_ID_BYTE_LEN 8
mlx4_en_get_phys_port_id(struct net_device * dev,struct netdev_phys_item_id * ppid)2626 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2627 				    struct netdev_phys_item_id *ppid)
2628 {
2629 	struct mlx4_en_priv *priv = netdev_priv(dev);
2630 	struct mlx4_dev *mdev = priv->mdev->dev;
2631 	int i;
2632 	u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2633 
2634 	if (!phys_port_id)
2635 		return -EOPNOTSUPP;
2636 
2637 	ppid->id_len = sizeof(phys_port_id);
2638 	for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2639 		ppid->id[i] =  phys_port_id & 0xff;
2640 		phys_port_id >>= 8;
2641 	}
2642 	return 0;
2643 }
2644 
mlx4_udp_tunnel_sync(struct net_device * dev,unsigned int table)2645 static int mlx4_udp_tunnel_sync(struct net_device *dev, unsigned int table)
2646 {
2647 	struct mlx4_en_priv *priv = netdev_priv(dev);
2648 	struct udp_tunnel_info ti;
2649 	int ret;
2650 
2651 	udp_tunnel_nic_get_port(dev, table, 0, &ti);
2652 	priv->vxlan_port = ti.port;
2653 
2654 	ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2655 	if (ret)
2656 		return ret;
2657 
2658 	return mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2659 				   VXLAN_STEER_BY_OUTER_MAC,
2660 				   !!priv->vxlan_port);
2661 }
2662 
2663 static const struct udp_tunnel_nic_info mlx4_udp_tunnels = {
2664 	.sync_table	= mlx4_udp_tunnel_sync,
2665 	.flags		= UDP_TUNNEL_NIC_INFO_MAY_SLEEP |
2666 			  UDP_TUNNEL_NIC_INFO_IPV4_ONLY,
2667 	.tables		= {
2668 		{ .n_entries = 1, .tunnel_types = UDP_TUNNEL_TYPE_VXLAN, },
2669 	},
2670 };
2671 
mlx4_en_features_check(struct sk_buff * skb,struct net_device * dev,netdev_features_t features)2672 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2673 						struct net_device *dev,
2674 						netdev_features_t features)
2675 {
2676 	features = vlan_features_check(skb, features);
2677 	features = vxlan_features_check(skb, features);
2678 
2679 	/* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2680 	 * support inner IPv6 checksums and segmentation so  we need to
2681 	 * strip that feature if this is an IPv6 encapsulated frame.
2682 	 */
2683 	if (skb->encapsulation &&
2684 	    (skb->ip_summed == CHECKSUM_PARTIAL)) {
2685 		struct mlx4_en_priv *priv = netdev_priv(dev);
2686 
2687 		if (!priv->vxlan_port ||
2688 		    (ip_hdr(skb)->version != 4) ||
2689 		    (udp_hdr(skb)->dest != priv->vxlan_port))
2690 			features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2691 	}
2692 
2693 	return features;
2694 }
2695 
mlx4_en_set_tx_maxrate(struct net_device * dev,int queue_index,u32 maxrate)2696 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2697 {
2698 	struct mlx4_en_priv *priv = netdev_priv(dev);
2699 	struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2700 	struct mlx4_update_qp_params params;
2701 	int err;
2702 
2703 	if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2704 		return -EOPNOTSUPP;
2705 
2706 	/* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2707 	if (maxrate >> 12) {
2708 		params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2709 		params.rate_val  = maxrate / 1000;
2710 	} else if (maxrate) {
2711 		params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2712 		params.rate_val  = maxrate;
2713 	} else { /* zero serves to revoke the QP rate-limitation */
2714 		params.rate_unit = 0;
2715 		params.rate_val  = 0;
2716 	}
2717 
2718 	err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2719 			     &params);
2720 	return err;
2721 }
2722 
mlx4_xdp_set(struct net_device * dev,struct bpf_prog * prog)2723 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2724 {
2725 	struct mlx4_en_priv *priv = netdev_priv(dev);
2726 	struct mlx4_en_dev *mdev = priv->mdev;
2727 	struct mlx4_en_port_profile new_prof;
2728 	struct bpf_prog *old_prog;
2729 	struct mlx4_en_priv *tmp;
2730 	int tx_changed = 0;
2731 	int xdp_ring_num;
2732 	int port_up = 0;
2733 	int err;
2734 	int i;
2735 
2736 	xdp_ring_num = prog ? priv->rx_ring_num : 0;
2737 
2738 	/* No need to reconfigure buffers when simply swapping the
2739 	 * program for a new one.
2740 	 */
2741 	if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2742 		if (prog)
2743 			bpf_prog_add(prog, priv->rx_ring_num - 1);
2744 
2745 		mutex_lock(&mdev->state_lock);
2746 		for (i = 0; i < priv->rx_ring_num; i++) {
2747 			old_prog = rcu_dereference_protected(
2748 					priv->rx_ring[i]->xdp_prog,
2749 					lockdep_is_held(&mdev->state_lock));
2750 			rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2751 			if (old_prog)
2752 				bpf_prog_put(old_prog);
2753 		}
2754 		mutex_unlock(&mdev->state_lock);
2755 		return 0;
2756 	}
2757 
2758 	if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2759 		return -EOPNOTSUPP;
2760 
2761 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2762 	if (!tmp)
2763 		return -ENOMEM;
2764 
2765 	if (prog)
2766 		bpf_prog_add(prog, priv->rx_ring_num - 1);
2767 
2768 	mutex_lock(&mdev->state_lock);
2769 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2770 	new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2771 
2772 	if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2773 		tx_changed = 1;
2774 		new_prof.tx_ring_num[TX] =
2775 			MAX_TX_RINGS - ALIGN(xdp_ring_num, priv->prof->num_up);
2776 		en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2777 	}
2778 
2779 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2780 	if (err) {
2781 		if (prog)
2782 			bpf_prog_sub(prog, priv->rx_ring_num - 1);
2783 		goto unlock_out;
2784 	}
2785 
2786 	if (priv->port_up) {
2787 		port_up = 1;
2788 		mlx4_en_stop_port(dev, 1);
2789 	}
2790 
2791 	mlx4_en_safe_replace_resources(priv, tmp);
2792 	if (tx_changed)
2793 		netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2794 
2795 	for (i = 0; i < priv->rx_ring_num; i++) {
2796 		old_prog = rcu_dereference_protected(
2797 					priv->rx_ring[i]->xdp_prog,
2798 					lockdep_is_held(&mdev->state_lock));
2799 		rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2800 		if (old_prog)
2801 			bpf_prog_put(old_prog);
2802 	}
2803 
2804 	if (port_up) {
2805 		err = mlx4_en_start_port(dev);
2806 		if (err) {
2807 			en_err(priv, "Failed starting port %d for XDP change\n",
2808 			       priv->port);
2809 			if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state))
2810 				queue_work(mdev->workqueue, &priv->restart_task);
2811 		}
2812 	}
2813 
2814 unlock_out:
2815 	mutex_unlock(&mdev->state_lock);
2816 	kfree(tmp);
2817 	return err;
2818 }
2819 
mlx4_xdp(struct net_device * dev,struct netdev_bpf * xdp)2820 static int mlx4_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2821 {
2822 	switch (xdp->command) {
2823 	case XDP_SETUP_PROG:
2824 		return mlx4_xdp_set(dev, xdp->prog);
2825 	default:
2826 		return -EINVAL;
2827 	}
2828 }
2829 
2830 static const struct net_device_ops mlx4_netdev_ops = {
2831 	.ndo_open		= mlx4_en_open,
2832 	.ndo_stop		= mlx4_en_close,
2833 	.ndo_start_xmit		= mlx4_en_xmit,
2834 	.ndo_select_queue	= mlx4_en_select_queue,
2835 	.ndo_get_stats64	= mlx4_en_get_stats64,
2836 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2837 	.ndo_set_mac_address	= mlx4_en_set_mac,
2838 	.ndo_validate_addr	= eth_validate_addr,
2839 	.ndo_change_mtu		= mlx4_en_change_mtu,
2840 	.ndo_do_ioctl		= mlx4_en_ioctl,
2841 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2842 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2843 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2844 	.ndo_set_features	= mlx4_en_set_features,
2845 	.ndo_fix_features	= mlx4_en_fix_features,
2846 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2847 #ifdef CONFIG_RFS_ACCEL
2848 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2849 #endif
2850 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2851 	.ndo_udp_tunnel_add	= udp_tunnel_nic_add_port,
2852 	.ndo_udp_tunnel_del	= udp_tunnel_nic_del_port,
2853 	.ndo_features_check	= mlx4_en_features_check,
2854 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2855 	.ndo_bpf		= mlx4_xdp,
2856 };
2857 
2858 static const struct net_device_ops mlx4_netdev_ops_master = {
2859 	.ndo_open		= mlx4_en_open,
2860 	.ndo_stop		= mlx4_en_close,
2861 	.ndo_start_xmit		= mlx4_en_xmit,
2862 	.ndo_select_queue	= mlx4_en_select_queue,
2863 	.ndo_get_stats64	= mlx4_en_get_stats64,
2864 	.ndo_set_rx_mode	= mlx4_en_set_rx_mode,
2865 	.ndo_set_mac_address	= mlx4_en_set_mac,
2866 	.ndo_validate_addr	= eth_validate_addr,
2867 	.ndo_change_mtu		= mlx4_en_change_mtu,
2868 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
2869 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
2870 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
2871 	.ndo_set_vf_mac		= mlx4_en_set_vf_mac,
2872 	.ndo_set_vf_vlan	= mlx4_en_set_vf_vlan,
2873 	.ndo_set_vf_rate	= mlx4_en_set_vf_rate,
2874 	.ndo_set_vf_spoofchk	= mlx4_en_set_vf_spoofchk,
2875 	.ndo_set_vf_link_state	= mlx4_en_set_vf_link_state,
2876 	.ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2877 	.ndo_get_vf_config	= mlx4_en_get_vf_config,
2878 	.ndo_set_features	= mlx4_en_set_features,
2879 	.ndo_fix_features	= mlx4_en_fix_features,
2880 	.ndo_setup_tc		= __mlx4_en_setup_tc,
2881 #ifdef CONFIG_RFS_ACCEL
2882 	.ndo_rx_flow_steer	= mlx4_en_filter_rfs,
2883 #endif
2884 	.ndo_get_phys_port_id	= mlx4_en_get_phys_port_id,
2885 	.ndo_udp_tunnel_add	= udp_tunnel_nic_add_port,
2886 	.ndo_udp_tunnel_del	= udp_tunnel_nic_del_port,
2887 	.ndo_features_check	= mlx4_en_features_check,
2888 	.ndo_set_tx_maxrate	= mlx4_en_set_tx_maxrate,
2889 	.ndo_bpf		= mlx4_xdp,
2890 };
2891 
2892 struct mlx4_en_bond {
2893 	struct work_struct work;
2894 	struct mlx4_en_priv *priv;
2895 	int is_bonded;
2896 	struct mlx4_port_map port_map;
2897 };
2898 
mlx4_en_bond_work(struct work_struct * work)2899 static void mlx4_en_bond_work(struct work_struct *work)
2900 {
2901 	struct mlx4_en_bond *bond = container_of(work,
2902 						     struct mlx4_en_bond,
2903 						     work);
2904 	int err = 0;
2905 	struct mlx4_dev *dev = bond->priv->mdev->dev;
2906 
2907 	if (bond->is_bonded) {
2908 		if (!mlx4_is_bonded(dev)) {
2909 			err = mlx4_bond(dev);
2910 			if (err)
2911 				en_err(bond->priv, "Fail to bond device\n");
2912 		}
2913 		if (!err) {
2914 			err = mlx4_port_map_set(dev, &bond->port_map);
2915 			if (err)
2916 				en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2917 				       bond->port_map.port1,
2918 				       bond->port_map.port2,
2919 				       err);
2920 		}
2921 	} else if (mlx4_is_bonded(dev)) {
2922 		err = mlx4_unbond(dev);
2923 		if (err)
2924 			en_err(bond->priv, "Fail to unbond device\n");
2925 	}
2926 	dev_put(bond->priv->dev);
2927 	kfree(bond);
2928 }
2929 
mlx4_en_queue_bond_work(struct mlx4_en_priv * priv,int is_bonded,u8 v2p_p1,u8 v2p_p2)2930 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2931 				   u8 v2p_p1, u8 v2p_p2)
2932 {
2933 	struct mlx4_en_bond *bond = NULL;
2934 
2935 	bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2936 	if (!bond)
2937 		return -ENOMEM;
2938 
2939 	INIT_WORK(&bond->work, mlx4_en_bond_work);
2940 	bond->priv = priv;
2941 	bond->is_bonded = is_bonded;
2942 	bond->port_map.port1 = v2p_p1;
2943 	bond->port_map.port2 = v2p_p2;
2944 	dev_hold(priv->dev);
2945 	queue_work(priv->mdev->workqueue, &bond->work);
2946 	return 0;
2947 }
2948 
mlx4_en_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)2949 int mlx4_en_netdev_event(struct notifier_block *this,
2950 			 unsigned long event, void *ptr)
2951 {
2952 	struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2953 	u8 port = 0;
2954 	struct mlx4_en_dev *mdev;
2955 	struct mlx4_dev *dev;
2956 	int i, num_eth_ports = 0;
2957 	bool do_bond = true;
2958 	struct mlx4_en_priv *priv;
2959 	u8 v2p_port1 = 0;
2960 	u8 v2p_port2 = 0;
2961 
2962 	if (!net_eq(dev_net(ndev), &init_net))
2963 		return NOTIFY_DONE;
2964 
2965 	mdev = container_of(this, struct mlx4_en_dev, nb);
2966 	dev = mdev->dev;
2967 
2968 	/* Go into this mode only when two network devices set on two ports
2969 	 * of the same mlx4 device are slaves of the same bonding master
2970 	 */
2971 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2972 		++num_eth_ports;
2973 		if (!port && (mdev->pndev[i] == ndev))
2974 			port = i;
2975 		mdev->upper[i] = mdev->pndev[i] ?
2976 			netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2977 		/* condition not met: network device is a slave */
2978 		if (!mdev->upper[i])
2979 			do_bond = false;
2980 		if (num_eth_ports < 2)
2981 			continue;
2982 		/* condition not met: same master */
2983 		if (mdev->upper[i] != mdev->upper[i-1])
2984 			do_bond = false;
2985 	}
2986 	/* condition not met: 2 salves */
2987 	do_bond = (num_eth_ports ==  2) ? do_bond : false;
2988 
2989 	/* handle only events that come with enough info */
2990 	if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
2991 		return NOTIFY_DONE;
2992 
2993 	priv = netdev_priv(ndev);
2994 	if (do_bond) {
2995 		struct netdev_notifier_bonding_info *notifier_info = ptr;
2996 		struct netdev_bonding_info *bonding_info =
2997 			&notifier_info->bonding_info;
2998 
2999 		/* required mode 1, 2 or 4 */
3000 		if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3001 		    (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
3002 		    (bonding_info->master.bond_mode != BOND_MODE_8023AD))
3003 			do_bond = false;
3004 
3005 		/* require exactly 2 slaves */
3006 		if (bonding_info->master.num_slaves != 2)
3007 			do_bond = false;
3008 
3009 		/* calc v2p */
3010 		if (do_bond) {
3011 			if (bonding_info->master.bond_mode ==
3012 			    BOND_MODE_ACTIVEBACKUP) {
3013 				/* in active-backup mode virtual ports are
3014 				 * mapped to the physical port of the active
3015 				 * slave */
3016 				if (bonding_info->slave.state ==
3017 				    BOND_STATE_BACKUP) {
3018 					if (port == 1) {
3019 						v2p_port1 = 2;
3020 						v2p_port2 = 2;
3021 					} else {
3022 						v2p_port1 = 1;
3023 						v2p_port2 = 1;
3024 					}
3025 				} else { /* BOND_STATE_ACTIVE */
3026 					if (port == 1) {
3027 						v2p_port1 = 1;
3028 						v2p_port2 = 1;
3029 					} else {
3030 						v2p_port1 = 2;
3031 						v2p_port2 = 2;
3032 					}
3033 				}
3034 			} else { /* Active-Active */
3035 				/* in active-active mode a virtual port is
3036 				 * mapped to the native physical port if and only
3037 				 * if the physical port is up */
3038 				__s8 link = bonding_info->slave.link;
3039 
3040 				if (port == 1)
3041 					v2p_port2 = 2;
3042 				else
3043 					v2p_port1 = 1;
3044 				if ((link == BOND_LINK_UP) ||
3045 				    (link == BOND_LINK_FAIL)) {
3046 					if (port == 1)
3047 						v2p_port1 = 1;
3048 					else
3049 						v2p_port2 = 2;
3050 				} else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3051 					if (port == 1)
3052 						v2p_port1 = 2;
3053 					else
3054 						v2p_port2 = 1;
3055 				}
3056 			}
3057 		}
3058 	}
3059 
3060 	mlx4_en_queue_bond_work(priv, do_bond,
3061 				v2p_port1, v2p_port2);
3062 
3063 	return NOTIFY_DONE;
3064 }
3065 
mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev * dev,struct mlx4_en_stats_bitmap * stats_bitmap,u8 rx_ppp,u8 rx_pause,u8 tx_ppp,u8 tx_pause)3066 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3067 				     struct mlx4_en_stats_bitmap *stats_bitmap,
3068 				     u8 rx_ppp, u8 rx_pause,
3069 				     u8 tx_ppp, u8 tx_pause)
3070 {
3071 	int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3072 
3073 	if (!mlx4_is_slave(dev) &&
3074 	    (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3075 		mutex_lock(&stats_bitmap->mutex);
3076 		bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3077 
3078 		if (rx_ppp)
3079 			bitmap_set(stats_bitmap->bitmap, last_i,
3080 				   NUM_FLOW_PRIORITY_STATS_RX);
3081 		last_i += NUM_FLOW_PRIORITY_STATS_RX;
3082 
3083 		if (rx_pause && !(rx_ppp))
3084 			bitmap_set(stats_bitmap->bitmap, last_i,
3085 				   NUM_FLOW_STATS_RX);
3086 		last_i += NUM_FLOW_STATS_RX;
3087 
3088 		if (tx_ppp)
3089 			bitmap_set(stats_bitmap->bitmap, last_i,
3090 				   NUM_FLOW_PRIORITY_STATS_TX);
3091 		last_i += NUM_FLOW_PRIORITY_STATS_TX;
3092 
3093 		if (tx_pause && !(tx_ppp))
3094 			bitmap_set(stats_bitmap->bitmap, last_i,
3095 				   NUM_FLOW_STATS_TX);
3096 		last_i += NUM_FLOW_STATS_TX;
3097 
3098 		mutex_unlock(&stats_bitmap->mutex);
3099 	}
3100 }
3101 
mlx4_en_set_stats_bitmap(struct mlx4_dev * dev,struct mlx4_en_stats_bitmap * stats_bitmap,u8 rx_ppp,u8 rx_pause,u8 tx_ppp,u8 tx_pause)3102 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3103 			      struct mlx4_en_stats_bitmap *stats_bitmap,
3104 			      u8 rx_ppp, u8 rx_pause,
3105 			      u8 tx_ppp, u8 tx_pause)
3106 {
3107 	int last_i = 0;
3108 
3109 	mutex_init(&stats_bitmap->mutex);
3110 	bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3111 
3112 	if (mlx4_is_slave(dev)) {
3113 		bitmap_set(stats_bitmap->bitmap, last_i +
3114 					 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3115 		bitmap_set(stats_bitmap->bitmap, last_i +
3116 					 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3117 		bitmap_set(stats_bitmap->bitmap, last_i +
3118 					 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3119 		bitmap_set(stats_bitmap->bitmap, last_i +
3120 					 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3121 		bitmap_set(stats_bitmap->bitmap, last_i +
3122 					 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3123 		bitmap_set(stats_bitmap->bitmap, last_i +
3124 					 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3125 	} else {
3126 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3127 	}
3128 	last_i += NUM_MAIN_STATS;
3129 
3130 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3131 	last_i += NUM_PORT_STATS;
3132 
3133 	if (mlx4_is_master(dev))
3134 		bitmap_set(stats_bitmap->bitmap, last_i,
3135 			   NUM_PF_STATS);
3136 	last_i += NUM_PF_STATS;
3137 
3138 	mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3139 					rx_ppp, rx_pause,
3140 					tx_ppp, tx_pause);
3141 	last_i += NUM_FLOW_STATS;
3142 
3143 	if (!mlx4_is_slave(dev))
3144 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3145 	last_i += NUM_PKT_STATS;
3146 
3147 	bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3148 	last_i += NUM_XDP_STATS;
3149 
3150 	if (!mlx4_is_slave(dev))
3151 		bitmap_set(stats_bitmap->bitmap, last_i, NUM_PHY_STATS);
3152 	last_i += NUM_PHY_STATS;
3153 }
3154 
mlx4_en_init_netdev(struct mlx4_en_dev * mdev,int port,struct mlx4_en_port_profile * prof)3155 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3156 			struct mlx4_en_port_profile *prof)
3157 {
3158 	struct net_device *dev;
3159 	struct mlx4_en_priv *priv;
3160 	int i, t;
3161 	int err;
3162 
3163 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3164 				 MAX_TX_RINGS, MAX_RX_RINGS);
3165 	if (dev == NULL)
3166 		return -ENOMEM;
3167 
3168 	netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3169 	netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3170 
3171 	SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3172 	dev->dev_port = port - 1;
3173 
3174 	/*
3175 	 * Initialize driver private data
3176 	 */
3177 
3178 	priv = netdev_priv(dev);
3179 	memset(priv, 0, sizeof(struct mlx4_en_priv));
3180 	priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3181 	spin_lock_init(&priv->stats_lock);
3182 	INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3183 	INIT_WORK(&priv->restart_task, mlx4_en_restart);
3184 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3185 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3186 	INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3187 #ifdef CONFIG_RFS_ACCEL
3188 	INIT_LIST_HEAD(&priv->filters);
3189 	spin_lock_init(&priv->filters_lock);
3190 #endif
3191 
3192 	priv->dev = dev;
3193 	priv->mdev = mdev;
3194 	priv->ddev = &mdev->pdev->dev;
3195 	priv->prof = prof;
3196 	priv->port = port;
3197 	priv->port_up = false;
3198 	priv->flags = prof->flags;
3199 	priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3200 	priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3201 			MLX4_WQE_CTRL_SOLICITED);
3202 	priv->num_tx_rings_p_up = mdev->profile.max_num_tx_rings_p_up;
3203 	priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3204 	netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3205 
3206 	for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3207 		priv->tx_ring_num[t] = prof->tx_ring_num[t];
3208 		if (!priv->tx_ring_num[t])
3209 			continue;
3210 
3211 		priv->tx_ring[t] = kcalloc(MAX_TX_RINGS,
3212 					   sizeof(struct mlx4_en_tx_ring *),
3213 					   GFP_KERNEL);
3214 		if (!priv->tx_ring[t]) {
3215 			err = -ENOMEM;
3216 			goto out;
3217 		}
3218 		priv->tx_cq[t] = kcalloc(MAX_TX_RINGS,
3219 					 sizeof(struct mlx4_en_cq *),
3220 					 GFP_KERNEL);
3221 		if (!priv->tx_cq[t]) {
3222 			err = -ENOMEM;
3223 			goto out;
3224 		}
3225 	}
3226 	priv->rx_ring_num = prof->rx_ring_num;
3227 	priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3228 	priv->cqe_size = mdev->dev->caps.cqe_size;
3229 	priv->mac_index = -1;
3230 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
3231 #ifdef CONFIG_MLX4_EN_DCB
3232 	if (!mlx4_is_slave(priv->mdev->dev)) {
3233 		u8 prio;
3234 
3235 		for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; ++prio) {
3236 			priv->ets.prio_tc[prio] = prio;
3237 			priv->ets.tc_tsa[prio]  = IEEE_8021QAZ_TSA_VENDOR;
3238 		}
3239 
3240 		priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3241 			DCB_CAP_DCBX_VER_IEEE;
3242 		priv->flags |= MLX4_EN_DCB_ENABLED;
3243 		priv->cee_config.pfc_state = false;
3244 
3245 		for (i = 0; i < MLX4_EN_NUM_UP_HIGH; i++)
3246 			priv->cee_config.dcb_pfc[i] = pfc_disabled;
3247 
3248 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3249 			dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3250 		} else {
3251 			en_info(priv, "enabling only PFC DCB ops\n");
3252 			dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3253 		}
3254 	}
3255 #endif
3256 
3257 	for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3258 		INIT_HLIST_HEAD(&priv->mac_hash[i]);
3259 
3260 	/* Query for default mac and max mtu */
3261 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3262 
3263 	if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3264 	    MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3265 		priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3266 
3267 	/* Set default MAC */
3268 	dev->addr_len = ETH_ALEN;
3269 	mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3270 	if (!is_valid_ether_addr(dev->dev_addr)) {
3271 		en_err(priv, "Port: %d, invalid mac burned: %pM, quitting\n",
3272 		       priv->port, dev->dev_addr);
3273 		err = -EINVAL;
3274 		goto out;
3275 	} else if (mlx4_is_slave(priv->mdev->dev) &&
3276 		   (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3277 		/* Random MAC was assigned in mlx4_slave_cap
3278 		 * in mlx4_core module
3279 		 */
3280 		dev->addr_assign_type |= NET_ADDR_RANDOM;
3281 		en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3282 	}
3283 
3284 	memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3285 
3286 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3287 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3288 	err = mlx4_en_alloc_resources(priv);
3289 	if (err)
3290 		goto out;
3291 
3292 	/* Initialize time stamping config */
3293 	priv->hwtstamp_config.flags = 0;
3294 	priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3295 	priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3296 
3297 	/* Allocate page for receive rings */
3298 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3299 				MLX4_EN_PAGE_SIZE);
3300 	if (err) {
3301 		en_err(priv, "Failed to allocate page for rx qps\n");
3302 		goto out;
3303 	}
3304 	priv->allocated = 1;
3305 
3306 	/*
3307 	 * Initialize netdev entry points
3308 	 */
3309 	if (mlx4_is_master(priv->mdev->dev))
3310 		dev->netdev_ops = &mlx4_netdev_ops_master;
3311 	else
3312 		dev->netdev_ops = &mlx4_netdev_ops;
3313 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3314 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3315 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3316 
3317 	dev->ethtool_ops = &mlx4_en_ethtool_ops;
3318 
3319 	/*
3320 	 * Set driver features
3321 	 */
3322 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3323 	if (mdev->LSO_support)
3324 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3325 
3326 	if (mdev->dev->caps.tunnel_offload_mode ==
3327 	    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3328 		dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3329 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3330 				    NETIF_F_GSO_PARTIAL;
3331 		dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3332 				    NETIF_F_GSO_UDP_TUNNEL_CSUM |
3333 				    NETIF_F_GSO_PARTIAL;
3334 		dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3335 		dev->hw_enc_features = NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3336 				       NETIF_F_RXCSUM |
3337 				       NETIF_F_TSO | NETIF_F_TSO6 |
3338 				       NETIF_F_GSO_UDP_TUNNEL |
3339 				       NETIF_F_GSO_UDP_TUNNEL_CSUM |
3340 				       NETIF_F_GSO_PARTIAL;
3341 
3342 		dev->udp_tunnel_nic_info = &mlx4_udp_tunnels;
3343 	}
3344 
3345 	dev->vlan_features = dev->hw_features;
3346 
3347 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3348 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3349 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3350 			NETIF_F_HW_VLAN_CTAG_FILTER;
3351 	dev->hw_features |= NETIF_F_LOOPBACK |
3352 			NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3353 
3354 	if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3355 		dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3356 			NETIF_F_HW_VLAN_STAG_FILTER;
3357 		dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3358 	}
3359 
3360 	if (mlx4_is_slave(mdev->dev)) {
3361 		bool vlan_offload_disabled;
3362 		int phv;
3363 
3364 		err = get_phv_bit(mdev->dev, port, &phv);
3365 		if (!err && phv) {
3366 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3367 			priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3368 		}
3369 		err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3370 							&vlan_offload_disabled);
3371 		if (!err && vlan_offload_disabled) {
3372 			dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3373 					      NETIF_F_HW_VLAN_CTAG_RX |
3374 					      NETIF_F_HW_VLAN_STAG_TX |
3375 					      NETIF_F_HW_VLAN_STAG_RX);
3376 			dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3377 					   NETIF_F_HW_VLAN_CTAG_RX |
3378 					   NETIF_F_HW_VLAN_STAG_TX |
3379 					   NETIF_F_HW_VLAN_STAG_RX);
3380 		}
3381 	} else {
3382 		if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3383 		    !(mdev->dev->caps.flags2 &
3384 		      MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3385 			dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3386 	}
3387 
3388 	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3389 		dev->hw_features |= NETIF_F_RXFCS;
3390 
3391 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3392 		dev->hw_features |= NETIF_F_RXALL;
3393 
3394 	if (mdev->dev->caps.steering_mode ==
3395 	    MLX4_STEERING_MODE_DEVICE_MANAGED &&
3396 	    mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3397 		dev->hw_features |= NETIF_F_NTUPLE;
3398 
3399 	if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3400 		dev->priv_flags |= IFF_UNICAST_FLT;
3401 
3402 	/* Setting a default hash function value */
3403 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3404 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3405 	} else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3406 		priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3407 	} else {
3408 		en_warn(priv,
3409 			"No RSS hash capabilities exposed, using Toeplitz\n");
3410 		priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3411 	}
3412 
3413 	/* MTU range: 68 - hw-specific max */
3414 	dev->min_mtu = ETH_MIN_MTU;
3415 	dev->max_mtu = priv->max_mtu;
3416 
3417 	mdev->pndev[port] = dev;
3418 	mdev->upper[port] = NULL;
3419 
3420 	netif_carrier_off(dev);
3421 	mlx4_en_set_default_moderation(priv);
3422 
3423 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3424 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3425 
3426 	mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3427 
3428 	/* Configure port */
3429 	mlx4_en_calc_rx_buf(dev);
3430 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3431 				    priv->rx_skb_size + ETH_FCS_LEN,
3432 				    prof->tx_pause, prof->tx_ppp,
3433 				    prof->rx_pause, prof->rx_ppp);
3434 	if (err) {
3435 		en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3436 		       priv->port, err);
3437 		goto out;
3438 	}
3439 
3440 	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3441 		err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3442 		if (err) {
3443 			en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3444 			       err);
3445 			goto out;
3446 		}
3447 	}
3448 
3449 	/* Init port */
3450 	en_warn(priv, "Initializing port\n");
3451 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
3452 	if (err) {
3453 		en_err(priv, "Failed Initializing port\n");
3454 		goto out;
3455 	}
3456 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3457 
3458 	/* Initialize time stamp mechanism */
3459 	if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3460 		mlx4_en_init_timestamp(mdev);
3461 
3462 	queue_delayed_work(mdev->workqueue, &priv->service_task,
3463 			   SERVICE_TASK_DELAY);
3464 
3465 	mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3466 				 mdev->profile.prof[priv->port].rx_ppp,
3467 				 mdev->profile.prof[priv->port].rx_pause,
3468 				 mdev->profile.prof[priv->port].tx_ppp,
3469 				 mdev->profile.prof[priv->port].tx_pause);
3470 
3471 	err = register_netdev(dev);
3472 	if (err) {
3473 		en_err(priv, "Netdev registration failed for port %d\n", port);
3474 		goto out;
3475 	}
3476 
3477 	priv->registered = 1;
3478 	devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3479 				  dev);
3480 
3481 	return 0;
3482 
3483 out:
3484 	mlx4_en_destroy_netdev(dev);
3485 	return err;
3486 }
3487 
mlx4_en_reset_config(struct net_device * dev,struct hwtstamp_config ts_config,netdev_features_t features)3488 int mlx4_en_reset_config(struct net_device *dev,
3489 			 struct hwtstamp_config ts_config,
3490 			 netdev_features_t features)
3491 {
3492 	struct mlx4_en_priv *priv = netdev_priv(dev);
3493 	struct mlx4_en_dev *mdev = priv->mdev;
3494 	struct mlx4_en_port_profile new_prof;
3495 	struct mlx4_en_priv *tmp;
3496 	int port_up = 0;
3497 	int err = 0;
3498 
3499 	if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3500 	    priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3501 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3502 	    !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3503 		return 0; /* Nothing to change */
3504 
3505 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3506 	    (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3507 	    (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3508 		en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3509 		return -EINVAL;
3510 	}
3511 
3512 	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3513 	if (!tmp)
3514 		return -ENOMEM;
3515 
3516 	mutex_lock(&mdev->state_lock);
3517 
3518 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3519 	memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3520 
3521 	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3522 	if (err)
3523 		goto out;
3524 
3525 	if (priv->port_up) {
3526 		port_up = 1;
3527 		mlx4_en_stop_port(dev, 1);
3528 	}
3529 
3530 	mlx4_en_safe_replace_resources(priv, tmp);
3531 
3532 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3533 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3534 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3535 		else
3536 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3537 	} else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3538 		/* RX time-stamping is OFF, update the RX vlan offload
3539 		 * to the latest wanted state
3540 		 */
3541 		if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3542 			dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3543 		else
3544 			dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3545 	}
3546 
3547 	if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3548 		if (features & NETIF_F_RXFCS)
3549 			dev->features |= NETIF_F_RXFCS;
3550 		else
3551 			dev->features &= ~NETIF_F_RXFCS;
3552 	}
3553 
3554 	/* RX vlan offload and RX time-stamping can't co-exist !
3555 	 * Regardless of the caller's choice,
3556 	 * Turn Off RX vlan offload in case of time-stamping is ON
3557 	 */
3558 	if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3559 		if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3560 			en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3561 		dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3562 	}
3563 
3564 	if (port_up) {
3565 		err = mlx4_en_start_port(dev);
3566 		if (err)
3567 			en_err(priv, "Failed starting port\n");
3568 	}
3569 
3570 	if (!err)
3571 		err = mlx4_en_moderation_update(priv);
3572 out:
3573 	mutex_unlock(&mdev->state_lock);
3574 	kfree(tmp);
3575 	if (!err)
3576 		netdev_features_change(dev);
3577 	return err;
3578 }
3579