xref: /linux/drivers/net/ethernet/mellanox/mlx4/en_netdev.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
1 /*
2  * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  */
33 
34 #include <linux/etherdevice.h>
35 #include <linux/tcp.h>
36 #include <linux/if_vlan.h>
37 #include <linux/delay.h>
38 #include <linux/slab.h>
39 
40 #include <linux/mlx4/driver.h>
41 #include <linux/mlx4/device.h>
42 #include <linux/mlx4/cmd.h>
43 #include <linux/mlx4/cq.h>
44 
45 #include "mlx4_en.h"
46 #include "en_port.h"
47 
48 static void mlx4_en_vlan_rx_add_vid(struct net_device *dev, unsigned short vid)
49 {
50 	struct mlx4_en_priv *priv = netdev_priv(dev);
51 	struct mlx4_en_dev *mdev = priv->mdev;
52 	int err;
53 	int idx;
54 
55 	en_dbg(HW, priv, "adding VLAN:%d\n", vid);
56 
57 	set_bit(vid, priv->active_vlans);
58 
59 	/* Add VID to port VLAN filter */
60 	mutex_lock(&mdev->state_lock);
61 	if (mdev->device_up && priv->port_up) {
62 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
63 		if (err)
64 			en_err(priv, "Failed configuring VLAN filter\n");
65 	}
66 	if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
67 		en_err(priv, "failed adding vlan %d\n", vid);
68 	mutex_unlock(&mdev->state_lock);
69 
70 }
71 
72 static void mlx4_en_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
73 {
74 	struct mlx4_en_priv *priv = netdev_priv(dev);
75 	struct mlx4_en_dev *mdev = priv->mdev;
76 	int err;
77 	int idx;
78 
79 	en_dbg(HW, priv, "Killing VID:%d\n", vid);
80 
81 	clear_bit(vid, priv->active_vlans);
82 
83 	/* Remove VID from port VLAN filter */
84 	mutex_lock(&mdev->state_lock);
85 	if (!mlx4_find_cached_vlan(mdev->dev, priv->port, vid, &idx))
86 		mlx4_unregister_vlan(mdev->dev, priv->port, idx);
87 	else
88 		en_err(priv, "could not find vid %d in cache\n", vid);
89 
90 	if (mdev->device_up && priv->port_up) {
91 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
92 		if (err)
93 			en_err(priv, "Failed configuring VLAN filter\n");
94 	}
95 	mutex_unlock(&mdev->state_lock);
96 }
97 
98 u64 mlx4_en_mac_to_u64(u8 *addr)
99 {
100 	u64 mac = 0;
101 	int i;
102 
103 	for (i = 0; i < ETH_ALEN; i++) {
104 		mac <<= 8;
105 		mac |= addr[i];
106 	}
107 	return mac;
108 }
109 
110 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
111 {
112 	struct mlx4_en_priv *priv = netdev_priv(dev);
113 	struct mlx4_en_dev *mdev = priv->mdev;
114 	struct sockaddr *saddr = addr;
115 
116 	if (!is_valid_ether_addr(saddr->sa_data))
117 		return -EADDRNOTAVAIL;
118 
119 	memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
120 	priv->mac = mlx4_en_mac_to_u64(dev->dev_addr);
121 	queue_work(mdev->workqueue, &priv->mac_task);
122 	return 0;
123 }
124 
125 static void mlx4_en_do_set_mac(struct work_struct *work)
126 {
127 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
128 						 mac_task);
129 	struct mlx4_en_dev *mdev = priv->mdev;
130 	int err = 0;
131 
132 	mutex_lock(&mdev->state_lock);
133 	if (priv->port_up) {
134 		/* Remove old MAC and insert the new one */
135 		err = mlx4_replace_mac(mdev->dev, priv->port,
136 				       priv->base_qpn, priv->mac, 0);
137 		if (err)
138 			en_err(priv, "Failed changing HW MAC address\n");
139 	} else
140 		en_dbg(HW, priv, "Port is down while "
141 				 "registering mac, exiting...\n");
142 
143 	mutex_unlock(&mdev->state_lock);
144 }
145 
146 static void mlx4_en_clear_list(struct net_device *dev)
147 {
148 	struct mlx4_en_priv *priv = netdev_priv(dev);
149 
150 	kfree(priv->mc_addrs);
151 	priv->mc_addrs_cnt = 0;
152 }
153 
154 static void mlx4_en_cache_mclist(struct net_device *dev)
155 {
156 	struct mlx4_en_priv *priv = netdev_priv(dev);
157 	struct netdev_hw_addr *ha;
158 	char *mc_addrs;
159 	int mc_addrs_cnt = netdev_mc_count(dev);
160 	int i;
161 
162 	mc_addrs = kmalloc(mc_addrs_cnt * ETH_ALEN, GFP_ATOMIC);
163 	if (!mc_addrs) {
164 		en_err(priv, "failed to allocate multicast list\n");
165 		return;
166 	}
167 	i = 0;
168 	netdev_for_each_mc_addr(ha, dev)
169 		memcpy(mc_addrs + i++ * ETH_ALEN, ha->addr, ETH_ALEN);
170 	priv->mc_addrs = mc_addrs;
171 	priv->mc_addrs_cnt = mc_addrs_cnt;
172 }
173 
174 
175 static void mlx4_en_set_multicast(struct net_device *dev)
176 {
177 	struct mlx4_en_priv *priv = netdev_priv(dev);
178 
179 	if (!priv->port_up)
180 		return;
181 
182 	queue_work(priv->mdev->workqueue, &priv->mcast_task);
183 }
184 
185 static void mlx4_en_do_set_multicast(struct work_struct *work)
186 {
187 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
188 						 mcast_task);
189 	struct mlx4_en_dev *mdev = priv->mdev;
190 	struct net_device *dev = priv->dev;
191 	u64 mcast_addr = 0;
192 	u8 mc_list[16] = {0};
193 	int err;
194 
195 	mutex_lock(&mdev->state_lock);
196 	if (!mdev->device_up) {
197 		en_dbg(HW, priv, "Card is not up, "
198 				 "ignoring multicast change.\n");
199 		goto out;
200 	}
201 	if (!priv->port_up) {
202 		en_dbg(HW, priv, "Port is down, "
203 				 "ignoring  multicast change.\n");
204 		goto out;
205 	}
206 
207 	/*
208 	 * Promsicuous mode: disable all filters
209 	 */
210 
211 	if (dev->flags & IFF_PROMISC) {
212 		if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
213 			if (netif_msg_rx_status(priv))
214 				en_warn(priv, "Entering promiscuous mode\n");
215 			priv->flags |= MLX4_EN_FLAG_PROMISC;
216 
217 			/* Enable promiscouos mode */
218 			if (!(mdev->dev->caps.flags &
219 						MLX4_DEV_CAP_FLAG_VEP_UC_STEER))
220 				err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port,
221 							     priv->base_qpn, 1);
222 			else
223 				err = mlx4_unicast_promisc_add(mdev->dev, priv->base_qpn,
224 							       priv->port);
225 			if (err)
226 				en_err(priv, "Failed enabling "
227 					     "promiscuous mode\n");
228 
229 			/* Disable port multicast filter (unconditionally) */
230 			err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
231 						  0, MLX4_MCAST_DISABLE);
232 			if (err)
233 				en_err(priv, "Failed disabling "
234 					     "multicast filter\n");
235 
236 			/* Add the default qp number as multicast promisc */
237 			if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
238 				err = mlx4_multicast_promisc_add(mdev->dev, priv->base_qpn,
239 								 priv->port);
240 				if (err)
241 					en_err(priv, "Failed entering multicast promisc mode\n");
242 				priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
243 			}
244 
245 			/* Disable port VLAN filter */
246 			err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
247 			if (err)
248 				en_err(priv, "Failed disabling VLAN filter\n");
249 		}
250 		goto out;
251 	}
252 
253 	/*
254 	 * Not in promiscuous mode
255 	 */
256 
257 	if (priv->flags & MLX4_EN_FLAG_PROMISC) {
258 		if (netif_msg_rx_status(priv))
259 			en_warn(priv, "Leaving promiscuous mode\n");
260 		priv->flags &= ~MLX4_EN_FLAG_PROMISC;
261 
262 		/* Disable promiscouos mode */
263 		if (!(mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER))
264 			err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port,
265 						     priv->base_qpn, 0);
266 		else
267 			err = mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
268 							  priv->port);
269 		if (err)
270 			en_err(priv, "Failed disabling promiscuous mode\n");
271 
272 		/* Disable Multicast promisc */
273 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
274 			err = mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
275 							    priv->port);
276 			if (err)
277 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
278 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
279 		}
280 
281 		/* Enable port VLAN filter */
282 		err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
283 		if (err)
284 			en_err(priv, "Failed enabling VLAN filter\n");
285 	}
286 
287 	/* Enable/disable the multicast filter according to IFF_ALLMULTI */
288 	if (dev->flags & IFF_ALLMULTI) {
289 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
290 					  0, MLX4_MCAST_DISABLE);
291 		if (err)
292 			en_err(priv, "Failed disabling multicast filter\n");
293 
294 		/* Add the default qp number as multicast promisc */
295 		if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
296 			err = mlx4_multicast_promisc_add(mdev->dev, priv->base_qpn,
297 							 priv->port);
298 			if (err)
299 				en_err(priv, "Failed entering multicast promisc mode\n");
300 			priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
301 		}
302 	} else {
303 		int i;
304 		/* Disable Multicast promisc */
305 		if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
306 			err = mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
307 							    priv->port);
308 			if (err)
309 				en_err(priv, "Failed disabling multicast promiscuous mode\n");
310 			priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
311 		}
312 
313 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
314 					  0, MLX4_MCAST_DISABLE);
315 		if (err)
316 			en_err(priv, "Failed disabling multicast filter\n");
317 
318 		/* Detach our qp from all the multicast addresses */
319 		for (i = 0; i < priv->mc_addrs_cnt; i++) {
320 			memcpy(&mc_list[10], priv->mc_addrs + i * ETH_ALEN, ETH_ALEN);
321 			mc_list[5] = priv->port;
322 			mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
323 					      mc_list, MLX4_PROT_ETH);
324 		}
325 		/* Flush mcast filter and init it with broadcast address */
326 		mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
327 				    1, MLX4_MCAST_CONFIG);
328 
329 		/* Update multicast list - we cache all addresses so they won't
330 		 * change while HW is updated holding the command semaphor */
331 		netif_tx_lock_bh(dev);
332 		mlx4_en_cache_mclist(dev);
333 		netif_tx_unlock_bh(dev);
334 		for (i = 0; i < priv->mc_addrs_cnt; i++) {
335 			mcast_addr =
336 			      mlx4_en_mac_to_u64(priv->mc_addrs + i * ETH_ALEN);
337 			memcpy(&mc_list[10], priv->mc_addrs + i * ETH_ALEN, ETH_ALEN);
338 			mc_list[5] = priv->port;
339 			mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp,
340 					      mc_list, 0, MLX4_PROT_ETH);
341 			mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
342 					    mcast_addr, 0, MLX4_MCAST_CONFIG);
343 		}
344 		err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
345 					  0, MLX4_MCAST_ENABLE);
346 		if (err)
347 			en_err(priv, "Failed enabling multicast filter\n");
348 	}
349 out:
350 	mutex_unlock(&mdev->state_lock);
351 }
352 
353 #ifdef CONFIG_NET_POLL_CONTROLLER
354 static void mlx4_en_netpoll(struct net_device *dev)
355 {
356 	struct mlx4_en_priv *priv = netdev_priv(dev);
357 	struct mlx4_en_cq *cq;
358 	unsigned long flags;
359 	int i;
360 
361 	for (i = 0; i < priv->rx_ring_num; i++) {
362 		cq = &priv->rx_cq[i];
363 		spin_lock_irqsave(&cq->lock, flags);
364 		napi_synchronize(&cq->napi);
365 		mlx4_en_process_rx_cq(dev, cq, 0);
366 		spin_unlock_irqrestore(&cq->lock, flags);
367 	}
368 }
369 #endif
370 
371 static void mlx4_en_tx_timeout(struct net_device *dev)
372 {
373 	struct mlx4_en_priv *priv = netdev_priv(dev);
374 	struct mlx4_en_dev *mdev = priv->mdev;
375 
376 	if (netif_msg_timer(priv))
377 		en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
378 
379 	priv->port_stats.tx_timeout++;
380 	en_dbg(DRV, priv, "Scheduling watchdog\n");
381 	queue_work(mdev->workqueue, &priv->watchdog_task);
382 }
383 
384 
385 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
386 {
387 	struct mlx4_en_priv *priv = netdev_priv(dev);
388 
389 	spin_lock_bh(&priv->stats_lock);
390 	memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
391 	spin_unlock_bh(&priv->stats_lock);
392 
393 	return &priv->ret_stats;
394 }
395 
396 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
397 {
398 	struct mlx4_en_cq *cq;
399 	int i;
400 
401 	/* If we haven't received a specific coalescing setting
402 	 * (module param), we set the moderation parameters as follows:
403 	 * - moder_cnt is set to the number of mtu sized packets to
404 	 *   satisfy our coelsing target.
405 	 * - moder_time is set to a fixed value.
406 	 */
407 	priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
408 	priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
409 	en_dbg(INTR, priv, "Default coalesing params for mtu:%d - "
410 			   "rx_frames:%d rx_usecs:%d\n",
411 		 priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
412 
413 	/* Setup cq moderation params */
414 	for (i = 0; i < priv->rx_ring_num; i++) {
415 		cq = &priv->rx_cq[i];
416 		cq->moder_cnt = priv->rx_frames;
417 		cq->moder_time = priv->rx_usecs;
418 		priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
419 		priv->last_moder_packets[i] = 0;
420 		priv->last_moder_bytes[i] = 0;
421 	}
422 
423 	for (i = 0; i < priv->tx_ring_num; i++) {
424 		cq = &priv->tx_cq[i];
425 		cq->moder_cnt = MLX4_EN_TX_COAL_PKTS;
426 		cq->moder_time = MLX4_EN_TX_COAL_TIME;
427 	}
428 
429 	/* Reset auto-moderation params */
430 	priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
431 	priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
432 	priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
433 	priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
434 	priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
435 	priv->adaptive_rx_coal = 1;
436 	priv->last_moder_jiffies = 0;
437 	priv->last_moder_tx_packets = 0;
438 }
439 
440 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
441 {
442 	unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
443 	struct mlx4_en_cq *cq;
444 	unsigned long packets;
445 	unsigned long rate;
446 	unsigned long avg_pkt_size;
447 	unsigned long rx_packets;
448 	unsigned long rx_bytes;
449 	unsigned long rx_pkt_diff;
450 	int moder_time;
451 	int ring, err;
452 
453 	if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
454 		return;
455 
456 	for (ring = 0; ring < priv->rx_ring_num; ring++) {
457 		spin_lock_bh(&priv->stats_lock);
458 		rx_packets = priv->rx_ring[ring].packets;
459 		rx_bytes = priv->rx_ring[ring].bytes;
460 		spin_unlock_bh(&priv->stats_lock);
461 
462 		rx_pkt_diff = ((unsigned long) (rx_packets -
463 				priv->last_moder_packets[ring]));
464 		packets = rx_pkt_diff;
465 		rate = packets * HZ / period;
466 		avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
467 				priv->last_moder_bytes[ring])) / packets : 0;
468 
469 		/* Apply auto-moderation only when packet rate
470 		 * exceeds a rate that it matters */
471 		if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
472 		    avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
473 			if (rate < priv->pkt_rate_low)
474 				moder_time = priv->rx_usecs_low;
475 			else if (rate > priv->pkt_rate_high)
476 				moder_time = priv->rx_usecs_high;
477 			else
478 				moder_time = (rate - priv->pkt_rate_low) *
479 					(priv->rx_usecs_high - priv->rx_usecs_low) /
480 					(priv->pkt_rate_high - priv->pkt_rate_low) +
481 					priv->rx_usecs_low;
482 		} else {
483 			moder_time = priv->rx_usecs_low;
484 		}
485 
486 		if (moder_time != priv->last_moder_time[ring]) {
487 			priv->last_moder_time[ring] = moder_time;
488 			cq = &priv->rx_cq[ring];
489 			cq->moder_time = moder_time;
490 			err = mlx4_en_set_cq_moder(priv, cq);
491 			if (err)
492 				en_err(priv, "Failed modifying moderation "
493 					     "for cq:%d\n", ring);
494 		}
495 		priv->last_moder_packets[ring] = rx_packets;
496 		priv->last_moder_bytes[ring] = rx_bytes;
497 	}
498 
499 	priv->last_moder_jiffies = jiffies;
500 }
501 
502 static void mlx4_en_do_get_stats(struct work_struct *work)
503 {
504 	struct delayed_work *delay = to_delayed_work(work);
505 	struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
506 						 stats_task);
507 	struct mlx4_en_dev *mdev = priv->mdev;
508 	int err;
509 
510 	err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
511 	if (err)
512 		en_dbg(HW, priv, "Could not update stats\n");
513 
514 	mutex_lock(&mdev->state_lock);
515 	if (mdev->device_up) {
516 		if (priv->port_up)
517 			mlx4_en_auto_moderation(priv);
518 
519 		queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
520 	}
521 	if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
522 		queue_work(mdev->workqueue, &priv->mac_task);
523 		mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
524 	}
525 	mutex_unlock(&mdev->state_lock);
526 }
527 
528 static void mlx4_en_linkstate(struct work_struct *work)
529 {
530 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
531 						 linkstate_task);
532 	struct mlx4_en_dev *mdev = priv->mdev;
533 	int linkstate = priv->link_state;
534 
535 	mutex_lock(&mdev->state_lock);
536 	/* If observable port state changed set carrier state and
537 	 * report to system log */
538 	if (priv->last_link_state != linkstate) {
539 		if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
540 			en_info(priv, "Link Down\n");
541 			netif_carrier_off(priv->dev);
542 		} else {
543 			en_info(priv, "Link Up\n");
544 			netif_carrier_on(priv->dev);
545 		}
546 	}
547 	priv->last_link_state = linkstate;
548 	mutex_unlock(&mdev->state_lock);
549 }
550 
551 
552 int mlx4_en_start_port(struct net_device *dev)
553 {
554 	struct mlx4_en_priv *priv = netdev_priv(dev);
555 	struct mlx4_en_dev *mdev = priv->mdev;
556 	struct mlx4_en_cq *cq;
557 	struct mlx4_en_tx_ring *tx_ring;
558 	int rx_index = 0;
559 	int tx_index = 0;
560 	int err = 0;
561 	int i;
562 	int j;
563 	u8 mc_list[16] = {0};
564 
565 	if (priv->port_up) {
566 		en_dbg(DRV, priv, "start port called while port already up\n");
567 		return 0;
568 	}
569 
570 	/* Calculate Rx buf size */
571 	dev->mtu = min(dev->mtu, priv->max_mtu);
572 	mlx4_en_calc_rx_buf(dev);
573 	en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
574 
575 	/* Configure rx cq's and rings */
576 	err = mlx4_en_activate_rx_rings(priv);
577 	if (err) {
578 		en_err(priv, "Failed to activate RX rings\n");
579 		return err;
580 	}
581 	for (i = 0; i < priv->rx_ring_num; i++) {
582 		cq = &priv->rx_cq[i];
583 
584 		err = mlx4_en_activate_cq(priv, cq, i);
585 		if (err) {
586 			en_err(priv, "Failed activating Rx CQ\n");
587 			goto cq_err;
588 		}
589 		for (j = 0; j < cq->size; j++)
590 			cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
591 		err = mlx4_en_set_cq_moder(priv, cq);
592 		if (err) {
593 			en_err(priv, "Failed setting cq moderation parameters");
594 			mlx4_en_deactivate_cq(priv, cq);
595 			goto cq_err;
596 		}
597 		mlx4_en_arm_cq(priv, cq);
598 		priv->rx_ring[i].cqn = cq->mcq.cqn;
599 		++rx_index;
600 	}
601 
602 	/* Set port mac number */
603 	en_dbg(DRV, priv, "Setting mac for port %d\n", priv->port);
604 	err = mlx4_register_mac(mdev->dev, priv->port,
605 				priv->mac, &priv->base_qpn, 0);
606 	if (err) {
607 		en_err(priv, "Failed setting port mac\n");
608 		goto cq_err;
609 	}
610 	mdev->mac_removed[priv->port] = 0;
611 
612 	err = mlx4_en_config_rss_steer(priv);
613 	if (err) {
614 		en_err(priv, "Failed configuring rss steering\n");
615 		goto mac_err;
616 	}
617 
618 	/* Configure tx cq's and rings */
619 	for (i = 0; i < priv->tx_ring_num; i++) {
620 		/* Configure cq */
621 		cq = &priv->tx_cq[i];
622 		err = mlx4_en_activate_cq(priv, cq, i);
623 		if (err) {
624 			en_err(priv, "Failed allocating Tx CQ\n");
625 			goto tx_err;
626 		}
627 		err = mlx4_en_set_cq_moder(priv, cq);
628 		if (err) {
629 			en_err(priv, "Failed setting cq moderation parameters");
630 			mlx4_en_deactivate_cq(priv, cq);
631 			goto tx_err;
632 		}
633 		en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
634 		cq->buf->wqe_index = cpu_to_be16(0xffff);
635 
636 		/* Configure ring */
637 		tx_ring = &priv->tx_ring[i];
638 		err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn);
639 		if (err) {
640 			en_err(priv, "Failed allocating Tx ring\n");
641 			mlx4_en_deactivate_cq(priv, cq);
642 			goto tx_err;
643 		}
644 		/* Set initial ownership of all Tx TXBBs to SW (1) */
645 		for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
646 			*((u32 *) (tx_ring->buf + j)) = 0xffffffff;
647 		++tx_index;
648 	}
649 
650 	/* Configure port */
651 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
652 				    priv->rx_skb_size + ETH_FCS_LEN,
653 				    priv->prof->tx_pause,
654 				    priv->prof->tx_ppp,
655 				    priv->prof->rx_pause,
656 				    priv->prof->rx_ppp);
657 	if (err) {
658 		en_err(priv, "Failed setting port general configurations "
659 			     "for port %d, with error %d\n", priv->port, err);
660 		goto tx_err;
661 	}
662 	/* Set default qp number */
663 	err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
664 	if (err) {
665 		en_err(priv, "Failed setting default qp numbers\n");
666 		goto tx_err;
667 	}
668 
669 	/* Init port */
670 	en_dbg(HW, priv, "Initializing port\n");
671 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
672 	if (err) {
673 		en_err(priv, "Failed Initializing port\n");
674 		goto tx_err;
675 	}
676 
677 	/* Attach rx QP to bradcast address */
678 	memset(&mc_list[10], 0xff, ETH_ALEN);
679 	mc_list[5] = priv->port;
680 	if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
681 				  0, MLX4_PROT_ETH))
682 		mlx4_warn(mdev, "Failed Attaching Broadcast\n");
683 
684 	/* Must redo promiscuous mode setup. */
685 	priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
686 
687 	/* Schedule multicast task to populate multicast list */
688 	queue_work(mdev->workqueue, &priv->mcast_task);
689 
690 	priv->port_up = true;
691 	netif_tx_start_all_queues(dev);
692 	return 0;
693 
694 tx_err:
695 	while (tx_index--) {
696 		mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[tx_index]);
697 		mlx4_en_deactivate_cq(priv, &priv->tx_cq[tx_index]);
698 	}
699 
700 	mlx4_en_release_rss_steer(priv);
701 mac_err:
702 	mlx4_unregister_mac(mdev->dev, priv->port, priv->base_qpn);
703 cq_err:
704 	while (rx_index--)
705 		mlx4_en_deactivate_cq(priv, &priv->rx_cq[rx_index]);
706 	for (i = 0; i < priv->rx_ring_num; i++)
707 		mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
708 
709 	return err; /* need to close devices */
710 }
711 
712 
713 void mlx4_en_stop_port(struct net_device *dev)
714 {
715 	struct mlx4_en_priv *priv = netdev_priv(dev);
716 	struct mlx4_en_dev *mdev = priv->mdev;
717 	int i;
718 	u8 mc_list[16] = {0};
719 
720 	if (!priv->port_up) {
721 		en_dbg(DRV, priv, "stop port called while port already down\n");
722 		return;
723 	}
724 
725 	/* Synchronize with tx routine */
726 	netif_tx_lock_bh(dev);
727 	netif_tx_stop_all_queues(dev);
728 	netif_tx_unlock_bh(dev);
729 
730 	/* Set port as not active */
731 	priv->port_up = false;
732 
733 	/* Detach All multicasts */
734 	memset(&mc_list[10], 0xff, ETH_ALEN);
735 	mc_list[5] = priv->port;
736 	mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
737 			      MLX4_PROT_ETH);
738 	for (i = 0; i < priv->mc_addrs_cnt; i++) {
739 		memcpy(&mc_list[10], priv->mc_addrs + i * ETH_ALEN, ETH_ALEN);
740 		mc_list[5] = priv->port;
741 		mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
742 				      mc_list, MLX4_PROT_ETH);
743 	}
744 	mlx4_en_clear_list(dev);
745 	/* Flush multicast filter */
746 	mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
747 
748 	/* Unregister Mac address for the port */
749 	mlx4_unregister_mac(mdev->dev, priv->port, priv->base_qpn);
750 	mdev->mac_removed[priv->port] = 1;
751 
752 	/* Free TX Rings */
753 	for (i = 0; i < priv->tx_ring_num; i++) {
754 		mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[i]);
755 		mlx4_en_deactivate_cq(priv, &priv->tx_cq[i]);
756 	}
757 	msleep(10);
758 
759 	for (i = 0; i < priv->tx_ring_num; i++)
760 		mlx4_en_free_tx_buf(dev, &priv->tx_ring[i]);
761 
762 	/* Free RSS qps */
763 	mlx4_en_release_rss_steer(priv);
764 
765 	/* Free RX Rings */
766 	for (i = 0; i < priv->rx_ring_num; i++) {
767 		mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
768 		while (test_bit(NAPI_STATE_SCHED, &priv->rx_cq[i].napi.state))
769 			msleep(1);
770 		mlx4_en_deactivate_cq(priv, &priv->rx_cq[i]);
771 	}
772 
773 	/* close port*/
774 	mlx4_CLOSE_PORT(mdev->dev, priv->port);
775 }
776 
777 static void mlx4_en_restart(struct work_struct *work)
778 {
779 	struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
780 						 watchdog_task);
781 	struct mlx4_en_dev *mdev = priv->mdev;
782 	struct net_device *dev = priv->dev;
783 
784 	en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
785 
786 	mutex_lock(&mdev->state_lock);
787 	if (priv->port_up) {
788 		mlx4_en_stop_port(dev);
789 		if (mlx4_en_start_port(dev))
790 			en_err(priv, "Failed restarting port %d\n", priv->port);
791 	}
792 	mutex_unlock(&mdev->state_lock);
793 }
794 
795 
796 static int mlx4_en_open(struct net_device *dev)
797 {
798 	struct mlx4_en_priv *priv = netdev_priv(dev);
799 	struct mlx4_en_dev *mdev = priv->mdev;
800 	int i;
801 	int err = 0;
802 
803 	mutex_lock(&mdev->state_lock);
804 
805 	if (!mdev->device_up) {
806 		en_err(priv, "Cannot open - device down/disabled\n");
807 		err = -EBUSY;
808 		goto out;
809 	}
810 
811 	/* Reset HW statistics and performance counters */
812 	if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
813 		en_dbg(HW, priv, "Failed dumping statistics\n");
814 
815 	memset(&priv->stats, 0, sizeof(priv->stats));
816 	memset(&priv->pstats, 0, sizeof(priv->pstats));
817 
818 	for (i = 0; i < priv->tx_ring_num; i++) {
819 		priv->tx_ring[i].bytes = 0;
820 		priv->tx_ring[i].packets = 0;
821 	}
822 	for (i = 0; i < priv->rx_ring_num; i++) {
823 		priv->rx_ring[i].bytes = 0;
824 		priv->rx_ring[i].packets = 0;
825 	}
826 
827 	err = mlx4_en_start_port(dev);
828 	if (err)
829 		en_err(priv, "Failed starting port:%d\n", priv->port);
830 
831 out:
832 	mutex_unlock(&mdev->state_lock);
833 	return err;
834 }
835 
836 
837 static int mlx4_en_close(struct net_device *dev)
838 {
839 	struct mlx4_en_priv *priv = netdev_priv(dev);
840 	struct mlx4_en_dev *mdev = priv->mdev;
841 
842 	en_dbg(IFDOWN, priv, "Close port called\n");
843 
844 	mutex_lock(&mdev->state_lock);
845 
846 	mlx4_en_stop_port(dev);
847 	netif_carrier_off(dev);
848 
849 	mutex_unlock(&mdev->state_lock);
850 	return 0;
851 }
852 
853 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
854 {
855 	int i;
856 
857 	for (i = 0; i < priv->tx_ring_num; i++) {
858 		if (priv->tx_ring[i].tx_info)
859 			mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
860 		if (priv->tx_cq[i].buf)
861 			mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
862 	}
863 
864 	for (i = 0; i < priv->rx_ring_num; i++) {
865 		if (priv->rx_ring[i].rx_info)
866 			mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i]);
867 		if (priv->rx_cq[i].buf)
868 			mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
869 	}
870 }
871 
872 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
873 {
874 	struct mlx4_en_port_profile *prof = priv->prof;
875 	int i;
876 	int base_tx_qpn, err;
877 
878 	err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &base_tx_qpn);
879 	if (err) {
880 		en_err(priv, "failed reserving range for TX rings\n");
881 		return err;
882 	}
883 
884 	/* Create tx Rings */
885 	for (i = 0; i < priv->tx_ring_num; i++) {
886 		if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
887 				      prof->tx_ring_size, i, TX))
888 			goto err;
889 
890 		if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i], base_tx_qpn + i,
891 					   prof->tx_ring_size, TXBB_SIZE))
892 			goto err;
893 	}
894 
895 	/* Create rx Rings */
896 	for (i = 0; i < priv->rx_ring_num; i++) {
897 		if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
898 				      prof->rx_ring_size, i, RX))
899 			goto err;
900 
901 		if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
902 					   prof->rx_ring_size, priv->stride))
903 			goto err;
904 	}
905 
906 	return 0;
907 
908 err:
909 	en_err(priv, "Failed to allocate NIC resources\n");
910 	mlx4_qp_release_range(priv->mdev->dev, base_tx_qpn, priv->tx_ring_num);
911 	return -ENOMEM;
912 }
913 
914 
915 void mlx4_en_destroy_netdev(struct net_device *dev)
916 {
917 	struct mlx4_en_priv *priv = netdev_priv(dev);
918 	struct mlx4_en_dev *mdev = priv->mdev;
919 
920 	en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
921 
922 	/* Unregister device - this will close the port if it was up */
923 	if (priv->registered)
924 		unregister_netdev(dev);
925 
926 	if (priv->allocated)
927 		mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
928 
929 	cancel_delayed_work(&priv->stats_task);
930 	/* flush any pending task for this netdev */
931 	flush_workqueue(mdev->workqueue);
932 
933 	/* Detach the netdev so tasks would not attempt to access it */
934 	mutex_lock(&mdev->state_lock);
935 	mdev->pndev[priv->port] = NULL;
936 	mutex_unlock(&mdev->state_lock);
937 
938 	mlx4_en_free_resources(priv);
939 	free_netdev(dev);
940 }
941 
942 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
943 {
944 	struct mlx4_en_priv *priv = netdev_priv(dev);
945 	struct mlx4_en_dev *mdev = priv->mdev;
946 	int err = 0;
947 
948 	en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
949 		 dev->mtu, new_mtu);
950 
951 	if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
952 		en_err(priv, "Bad MTU size:%d.\n", new_mtu);
953 		return -EPERM;
954 	}
955 	dev->mtu = new_mtu;
956 
957 	if (netif_running(dev)) {
958 		mutex_lock(&mdev->state_lock);
959 		if (!mdev->device_up) {
960 			/* NIC is probably restarting - let watchdog task reset
961 			 * the port */
962 			en_dbg(DRV, priv, "Change MTU called with card down!?\n");
963 		} else {
964 			mlx4_en_stop_port(dev);
965 			err = mlx4_en_start_port(dev);
966 			if (err) {
967 				en_err(priv, "Failed restarting port:%d\n",
968 					 priv->port);
969 				queue_work(mdev->workqueue, &priv->watchdog_task);
970 			}
971 		}
972 		mutex_unlock(&mdev->state_lock);
973 	}
974 	return 0;
975 }
976 
977 static const struct net_device_ops mlx4_netdev_ops = {
978 	.ndo_open		= mlx4_en_open,
979 	.ndo_stop		= mlx4_en_close,
980 	.ndo_start_xmit		= mlx4_en_xmit,
981 	.ndo_select_queue	= mlx4_en_select_queue,
982 	.ndo_get_stats		= mlx4_en_get_stats,
983 	.ndo_set_rx_mode	= mlx4_en_set_multicast,
984 	.ndo_set_mac_address	= mlx4_en_set_mac,
985 	.ndo_validate_addr	= eth_validate_addr,
986 	.ndo_change_mtu		= mlx4_en_change_mtu,
987 	.ndo_tx_timeout		= mlx4_en_tx_timeout,
988 	.ndo_vlan_rx_add_vid	= mlx4_en_vlan_rx_add_vid,
989 	.ndo_vlan_rx_kill_vid	= mlx4_en_vlan_rx_kill_vid,
990 #ifdef CONFIG_NET_POLL_CONTROLLER
991 	.ndo_poll_controller	= mlx4_en_netpoll,
992 #endif
993 };
994 
995 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
996 			struct mlx4_en_port_profile *prof)
997 {
998 	struct net_device *dev;
999 	struct mlx4_en_priv *priv;
1000 	int i;
1001 	int err;
1002 
1003 	dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
1004 	    prof->tx_ring_num, prof->rx_ring_num);
1005 	if (dev == NULL) {
1006 		mlx4_err(mdev, "Net device allocation failed\n");
1007 		return -ENOMEM;
1008 	}
1009 
1010 	SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
1011 	dev->dev_id =  port - 1;
1012 
1013 	/*
1014 	 * Initialize driver private data
1015 	 */
1016 
1017 	priv = netdev_priv(dev);
1018 	memset(priv, 0, sizeof(struct mlx4_en_priv));
1019 	priv->dev = dev;
1020 	priv->mdev = mdev;
1021 	priv->prof = prof;
1022 	priv->port = port;
1023 	priv->port_up = false;
1024 	priv->flags = prof->flags;
1025 	priv->tx_ring_num = prof->tx_ring_num;
1026 	priv->rx_ring_num = prof->rx_ring_num;
1027 	priv->mac_index = -1;
1028 	priv->msg_enable = MLX4_EN_MSG_LEVEL;
1029 	spin_lock_init(&priv->stats_lock);
1030 	INIT_WORK(&priv->mcast_task, mlx4_en_do_set_multicast);
1031 	INIT_WORK(&priv->mac_task, mlx4_en_do_set_mac);
1032 	INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
1033 	INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
1034 	INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
1035 
1036 	/* Query for default mac and max mtu */
1037 	priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
1038 	priv->mac = mdev->dev->caps.def_mac[priv->port];
1039 	if (ILLEGAL_MAC(priv->mac)) {
1040 		en_err(priv, "Port: %d, invalid mac burned: 0x%llx, quiting\n",
1041 			 priv->port, priv->mac);
1042 		err = -EINVAL;
1043 		goto out;
1044 	}
1045 
1046 	priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
1047 					  DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
1048 	err = mlx4_en_alloc_resources(priv);
1049 	if (err)
1050 		goto out;
1051 
1052 	/* Allocate page for receive rings */
1053 	err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
1054 				MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
1055 	if (err) {
1056 		en_err(priv, "Failed to allocate page for rx qps\n");
1057 		goto out;
1058 	}
1059 	priv->allocated = 1;
1060 
1061 	/*
1062 	 * Initialize netdev entry points
1063 	 */
1064 	dev->netdev_ops = &mlx4_netdev_ops;
1065 	dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
1066 	netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
1067 	netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
1068 
1069 	SET_ETHTOOL_OPS(dev, &mlx4_en_ethtool_ops);
1070 
1071 	/* Set defualt MAC */
1072 	dev->addr_len = ETH_ALEN;
1073 	for (i = 0; i < ETH_ALEN; i++) {
1074 		dev->dev_addr[ETH_ALEN - 1 - i] = (u8) (priv->mac >> (8 * i));
1075 		dev->perm_addr[ETH_ALEN - 1 - i] = (u8) (priv->mac >> (8 * i));
1076 	}
1077 
1078 	/*
1079 	 * Set driver features
1080 	 */
1081 	dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1082 	if (mdev->LSO_support)
1083 		dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
1084 
1085 	dev->vlan_features = dev->hw_features;
1086 
1087 	dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
1088 	dev->features = dev->hw_features | NETIF_F_HIGHDMA |
1089 			NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
1090 			NETIF_F_HW_VLAN_FILTER;
1091 
1092 	mdev->pndev[port] = dev;
1093 
1094 	netif_carrier_off(dev);
1095 	err = register_netdev(dev);
1096 	if (err) {
1097 		en_err(priv, "Netdev registration failed for port %d\n", port);
1098 		goto out;
1099 	}
1100 	priv->registered = 1;
1101 
1102 	en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
1103 	en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
1104 
1105 	/* Configure port */
1106 	err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1107 				    MLX4_EN_MIN_MTU,
1108 				    0, 0, 0, 0);
1109 	if (err) {
1110 		en_err(priv, "Failed setting port general configurations "
1111 		       "for port %d, with error %d\n", priv->port, err);
1112 		goto out;
1113 	}
1114 
1115 	/* Init port */
1116 	en_warn(priv, "Initializing port\n");
1117 	err = mlx4_INIT_PORT(mdev->dev, priv->port);
1118 	if (err) {
1119 		en_err(priv, "Failed Initializing port\n");
1120 		goto out;
1121 	}
1122 	mlx4_en_set_default_moderation(priv);
1123 	queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1124 	return 0;
1125 
1126 out:
1127 	mlx4_en_destroy_netdev(dev);
1128 	return err;
1129 }
1130 
1131