2 * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
34 #include <linux/etherdevice.h>
35 #include <linux/tcp.h>
36 #include <linux/if_vlan.h>
37 #include <linux/delay.h>
38 #include <linux/slab.h>
39 #include <linux/hash.h>
42 #include <linux/mlx4/driver.h>
43 #include <linux/mlx4/device.h>
44 #include <linux/mlx4/cmd.h>
45 #include <linux/mlx4/cq.h>
50 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
52 struct mlx4_en_priv *priv = netdev_priv(dev);
54 unsigned int offset = 0;
56 if (up && up != MLX4_EN_NUM_UP)
59 netdev_set_num_tc(dev, up);
61 /* Partition Tx queues evenly amongst UP's */
62 for (i = 0; i < up; i++) {
63 netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
64 offset += priv->num_tx_rings_p_up;
70 #ifdef CONFIG_RFS_ACCEL
72 struct mlx4_en_filter {
73 struct list_head next;
74 struct work_struct work;
82 struct mlx4_en_priv *priv;
83 u32 flow_id; /* RFS infrastructure id */
84 int id; /* mlx4_en driver id */
85 u64 reg_id; /* Flow steering API id */
86 u8 activated; /* Used to prevent expiry before filter
89 struct hlist_node filter_chain;
92 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
94 static void mlx4_en_filter_work(struct work_struct *work)
96 struct mlx4_en_filter *filter = container_of(work,
97 struct mlx4_en_filter,
99 struct mlx4_en_priv *priv = filter->priv;
100 struct mlx4_spec_list spec_tcp = {
101 .id = MLX4_NET_TRANS_RULE_ID_TCP,
104 .dst_port = filter->dst_port,
105 .dst_port_msk = (__force __be16)-1,
106 .src_port = filter->src_port,
107 .src_port_msk = (__force __be16)-1,
111 struct mlx4_spec_list spec_ip = {
112 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
115 .dst_ip = filter->dst_ip,
116 .dst_ip_msk = (__force __be32)-1,
117 .src_ip = filter->src_ip,
118 .src_ip_msk = (__force __be32)-1,
122 struct mlx4_spec_list spec_eth = {
123 .id = MLX4_NET_TRANS_RULE_ID_ETH,
125 struct mlx4_net_trans_rule rule = {
126 .list = LIST_HEAD_INIT(rule.list),
127 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
130 .promisc_mode = MLX4_FS_PROMISC_NONE,
132 .priority = MLX4_DOMAIN_RFS,
135 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
137 list_add_tail(&spec_eth.list, &rule.list);
138 list_add_tail(&spec_ip.list, &rule.list);
139 list_add_tail(&spec_tcp.list, &rule.list);
141 rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
142 memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
143 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
145 filter->activated = 0;
147 if (filter->reg_id) {
148 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
149 if (rc && rc != -ENOENT)
150 en_err(priv, "Error detaching flow. rc = %d\n", rc);
153 rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
155 en_err(priv, "Error attaching flow. err = %d\n", rc);
157 mlx4_en_filter_rfs_expire(priv);
159 filter->activated = 1;
162 static inline struct hlist_head *
163 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
164 __be16 src_port, __be16 dst_port)
169 l = (__force unsigned long)src_port |
170 ((__force unsigned long)dst_port << 2);
171 l ^= (__force unsigned long)(src_ip ^ dst_ip);
173 bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
175 return &priv->filter_hash[bucket_idx];
178 static struct mlx4_en_filter *
179 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
180 __be32 dst_ip, __be16 src_port, __be16 dst_port,
183 struct mlx4_en_filter *filter = NULL;
185 filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
190 filter->rxq_index = rxq_index;
191 INIT_WORK(&filter->work, mlx4_en_filter_work);
193 filter->src_ip = src_ip;
194 filter->dst_ip = dst_ip;
195 filter->src_port = src_port;
196 filter->dst_port = dst_port;
198 filter->flow_id = flow_id;
200 filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
202 list_add_tail(&filter->next, &priv->filters);
203 hlist_add_head(&filter->filter_chain,
204 filter_hash_bucket(priv, src_ip, dst_ip, src_port,
210 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
212 struct mlx4_en_priv *priv = filter->priv;
215 list_del(&filter->next);
217 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
218 if (rc && rc != -ENOENT)
219 en_err(priv, "Error detaching flow. rc = %d\n", rc);
224 static inline struct mlx4_en_filter *
225 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
226 __be16 src_port, __be16 dst_port)
228 struct mlx4_en_filter *filter;
229 struct mlx4_en_filter *ret = NULL;
231 hlist_for_each_entry(filter,
232 filter_hash_bucket(priv, src_ip, dst_ip,
235 if (filter->src_ip == src_ip &&
236 filter->dst_ip == dst_ip &&
237 filter->src_port == src_port &&
238 filter->dst_port == dst_port) {
248 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
249 u16 rxq_index, u32 flow_id)
251 struct mlx4_en_priv *priv = netdev_priv(net_dev);
252 struct mlx4_en_filter *filter;
253 const struct iphdr *ip;
259 int nhoff = skb_network_offset(skb);
262 if (skb->protocol != htons(ETH_P_IP))
263 return -EPROTONOSUPPORT;
265 ip = (const struct iphdr *)(skb->data + nhoff);
266 if (ip_is_fragment(ip))
267 return -EPROTONOSUPPORT;
269 ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
276 if (ip->protocol != IPPROTO_TCP)
277 return -EPROTONOSUPPORT;
279 spin_lock_bh(&priv->filters_lock);
280 filter = mlx4_en_filter_find(priv, src_ip, dst_ip, src_port, dst_port);
282 if (filter->rxq_index == rxq_index)
285 filter->rxq_index = rxq_index;
287 filter = mlx4_en_filter_alloc(priv, rxq_index,
289 src_port, dst_port, flow_id);
296 queue_work(priv->mdev->workqueue, &filter->work);
301 spin_unlock_bh(&priv->filters_lock);
306 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv,
307 struct mlx4_en_rx_ring *rx_ring)
309 struct mlx4_en_filter *filter, *tmp;
312 spin_lock_bh(&priv->filters_lock);
313 list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
314 list_move(&filter->next, &del_list);
315 hlist_del(&filter->filter_chain);
317 spin_unlock_bh(&priv->filters_lock);
319 list_for_each_entry_safe(filter, tmp, &del_list, next) {
320 cancel_work_sync(&filter->work);
321 mlx4_en_filter_free(filter);
325 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
327 struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
331 spin_lock_bh(&priv->filters_lock);
332 list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
333 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
336 if (filter->activated &&
337 !work_pending(&filter->work) &&
338 rps_may_expire_flow(priv->dev,
339 filter->rxq_index, filter->flow_id,
341 list_move(&filter->next, &del_list);
342 hlist_del(&filter->filter_chain);
344 last_filter = filter;
349 if (last_filter && (&last_filter->next != priv->filters.next))
350 list_move(&priv->filters, &last_filter->next);
352 spin_unlock_bh(&priv->filters_lock);
354 list_for_each_entry_safe(filter, tmp, &del_list, next)
355 mlx4_en_filter_free(filter);
359 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev, unsigned short vid)
361 struct mlx4_en_priv *priv = netdev_priv(dev);
362 struct mlx4_en_dev *mdev = priv->mdev;
366 en_dbg(HW, priv, "adding VLAN:%d\n", vid);
368 set_bit(vid, priv->active_vlans);
370 /* Add VID to port VLAN filter */
371 mutex_lock(&mdev->state_lock);
372 if (mdev->device_up && priv->port_up) {
373 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
375 en_err(priv, "Failed configuring VLAN filter\n");
377 if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
378 en_err(priv, "failed adding vlan %d\n", vid);
379 mutex_unlock(&mdev->state_lock);
384 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
386 struct mlx4_en_priv *priv = netdev_priv(dev);
387 struct mlx4_en_dev *mdev = priv->mdev;
391 en_dbg(HW, priv, "Killing VID:%d\n", vid);
393 clear_bit(vid, priv->active_vlans);
395 /* Remove VID from port VLAN filter */
396 mutex_lock(&mdev->state_lock);
397 if (!mlx4_find_cached_vlan(mdev->dev, priv->port, vid, &idx))
398 mlx4_unregister_vlan(mdev->dev, priv->port, idx);
400 en_err(priv, "could not find vid %d in cache\n", vid);
402 if (mdev->device_up && priv->port_up) {
403 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
405 en_err(priv, "Failed configuring VLAN filter\n");
407 mutex_unlock(&mdev->state_lock);
412 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
415 for (i = ETH_ALEN - 1; i; --i) {
416 dst_mac[i] = src_mac & 0xff;
419 memset(&dst_mac[ETH_ALEN], 0, 2);
422 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
423 unsigned char *mac, int *qpn, u64 *reg_id)
425 struct mlx4_en_dev *mdev = priv->mdev;
426 struct mlx4_dev *dev = mdev->dev;
429 switch (dev->caps.steering_mode) {
430 case MLX4_STEERING_MODE_B0: {
435 memcpy(&gid[10], mac, ETH_ALEN);
438 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
441 case MLX4_STEERING_MODE_DEVICE_MANAGED: {
442 struct mlx4_spec_list spec_eth = { {NULL} };
443 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
445 struct mlx4_net_trans_rule rule = {
446 .queue_mode = MLX4_NET_TRANS_Q_FIFO,
449 .promisc_mode = MLX4_FS_PROMISC_NONE,
450 .priority = MLX4_DOMAIN_NIC,
453 rule.port = priv->port;
455 INIT_LIST_HEAD(&rule.list);
457 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
458 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
459 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
460 list_add_tail(&spec_eth.list, &rule.list);
462 err = mlx4_flow_attach(dev, &rule, reg_id);
469 en_warn(priv, "Failed Attaching Unicast\n");
474 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
475 unsigned char *mac, int qpn, u64 reg_id)
477 struct mlx4_en_dev *mdev = priv->mdev;
478 struct mlx4_dev *dev = mdev->dev;
480 switch (dev->caps.steering_mode) {
481 case MLX4_STEERING_MODE_B0: {
486 memcpy(&gid[10], mac, ETH_ALEN);
489 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
492 case MLX4_STEERING_MODE_DEVICE_MANAGED: {
493 mlx4_flow_detach(dev, reg_id);
497 en_err(priv, "Invalid steering mode.\n");
501 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
503 struct mlx4_en_dev *mdev = priv->mdev;
504 struct mlx4_dev *dev = mdev->dev;
505 struct mlx4_mac_entry *entry;
509 int *qpn = &priv->base_qpn;
510 u64 mac = mlx4_en_mac_to_u64(priv->dev->dev_addr);
512 en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
513 priv->dev->dev_addr);
514 index = mlx4_register_mac(dev, priv->port, mac);
517 en_err(priv, "Failed adding MAC: %pM\n",
518 priv->dev->dev_addr);
522 if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
523 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
524 *qpn = base_qpn + index;
528 err = mlx4_qp_reserve_range(dev, 1, 1, qpn);
529 en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
531 en_err(priv, "Failed to reserve qp for mac registration\n");
535 err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, ®_id);
539 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
544 memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
545 entry->reg_id = reg_id;
547 hlist_add_head_rcu(&entry->hlist,
548 &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
553 mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
556 mlx4_qp_release_range(dev, *qpn, 1);
559 mlx4_unregister_mac(dev, priv->port, mac);
563 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
565 struct mlx4_en_dev *mdev = priv->mdev;
566 struct mlx4_dev *dev = mdev->dev;
567 int qpn = priv->base_qpn;
568 u64 mac = mlx4_en_mac_to_u64(priv->dev->dev_addr);
570 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
571 priv->dev->dev_addr);
572 mlx4_unregister_mac(dev, priv->port, mac);
574 if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
575 struct mlx4_mac_entry *entry;
576 struct hlist_node *tmp;
577 struct hlist_head *bucket;
578 unsigned int mac_hash;
580 mac_hash = priv->dev->dev_addr[MLX4_EN_MAC_HASH_IDX];
581 bucket = &priv->mac_hash[mac_hash];
582 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
583 if (ether_addr_equal_64bits(entry->mac,
584 priv->dev->dev_addr)) {
585 en_dbg(DRV, priv, "Releasing qp: port %d, MAC %pM, qpn %d\n",
586 priv->port, priv->dev->dev_addr, qpn);
587 mlx4_en_uc_steer_release(priv, entry->mac,
589 mlx4_qp_release_range(dev, qpn, 1);
591 hlist_del_rcu(&entry->hlist);
592 kfree_rcu(entry, rcu);
599 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
600 unsigned char *new_mac, unsigned char *prev_mac)
602 struct mlx4_en_dev *mdev = priv->mdev;
603 struct mlx4_dev *dev = mdev->dev;
605 u64 new_mac_u64 = mlx4_en_mac_to_u64(new_mac);
607 if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
608 struct hlist_head *bucket;
609 unsigned int mac_hash;
610 struct mlx4_mac_entry *entry;
611 struct hlist_node *tmp;
612 u64 prev_mac_u64 = mlx4_en_mac_to_u64(prev_mac);
614 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
615 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
616 if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
617 mlx4_en_uc_steer_release(priv, entry->mac,
619 mlx4_unregister_mac(dev, priv->port,
621 hlist_del_rcu(&entry->hlist);
623 memcpy(entry->mac, new_mac, ETH_ALEN);
625 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
626 hlist_add_head_rcu(&entry->hlist,
627 &priv->mac_hash[mac_hash]);
628 mlx4_register_mac(dev, priv->port, new_mac_u64);
629 err = mlx4_en_uc_steer_add(priv, new_mac,
638 return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
641 u64 mlx4_en_mac_to_u64(u8 *addr)
646 for (i = 0; i < ETH_ALEN; i++) {
653 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
655 struct mlx4_en_priv *priv = netdev_priv(dev);
656 struct mlx4_en_dev *mdev = priv->mdev;
657 struct sockaddr *saddr = addr;
659 if (!is_valid_ether_addr(saddr->sa_data))
660 return -EADDRNOTAVAIL;
662 memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
663 queue_work(mdev->workqueue, &priv->mac_task);
667 static void mlx4_en_do_set_mac(struct work_struct *work)
669 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
671 struct mlx4_en_dev *mdev = priv->mdev;
674 mutex_lock(&mdev->state_lock);
676 /* Remove old MAC and insert the new one */
677 err = mlx4_en_replace_mac(priv, priv->base_qpn,
678 priv->dev->dev_addr, priv->prev_mac);
680 en_err(priv, "Failed changing HW MAC address\n");
681 memcpy(priv->prev_mac, priv->dev->dev_addr,
682 sizeof(priv->prev_mac));
684 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
686 mutex_unlock(&mdev->state_lock);
689 static void mlx4_en_clear_list(struct net_device *dev)
691 struct mlx4_en_priv *priv = netdev_priv(dev);
692 struct mlx4_en_mc_list *tmp, *mc_to_del;
694 list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
695 list_del(&mc_to_del->list);
700 static void mlx4_en_cache_mclist(struct net_device *dev)
702 struct mlx4_en_priv *priv = netdev_priv(dev);
703 struct netdev_hw_addr *ha;
704 struct mlx4_en_mc_list *tmp;
706 mlx4_en_clear_list(dev);
707 netdev_for_each_mc_addr(ha, dev) {
708 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
710 mlx4_en_clear_list(dev);
713 memcpy(tmp->addr, ha->addr, ETH_ALEN);
714 list_add_tail(&tmp->list, &priv->mc_list);
718 static void update_mclist_flags(struct mlx4_en_priv *priv,
719 struct list_head *dst,
720 struct list_head *src)
722 struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
725 /* Find all the entries that should be removed from dst,
726 * These are the entries that are not found in src
728 list_for_each_entry(dst_tmp, dst, list) {
730 list_for_each_entry(src_tmp, src, list) {
731 if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
737 dst_tmp->action = MCLIST_REM;
740 /* Add entries that exist in src but not in dst
741 * mark them as need to add
743 list_for_each_entry(src_tmp, src, list) {
745 list_for_each_entry(dst_tmp, dst, list) {
746 if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
747 dst_tmp->action = MCLIST_NONE;
753 new_mc = kmemdup(src_tmp,
754 sizeof(struct mlx4_en_mc_list),
759 new_mc->action = MCLIST_ADD;
760 list_add_tail(&new_mc->list, dst);
765 static void mlx4_en_set_rx_mode(struct net_device *dev)
767 struct mlx4_en_priv *priv = netdev_priv(dev);
772 queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
775 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
776 struct mlx4_en_dev *mdev)
780 if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
781 if (netif_msg_rx_status(priv))
782 en_warn(priv, "Entering promiscuous mode\n");
783 priv->flags |= MLX4_EN_FLAG_PROMISC;
785 /* Enable promiscouos mode */
786 switch (mdev->dev->caps.steering_mode) {
787 case MLX4_STEERING_MODE_DEVICE_MANAGED:
788 err = mlx4_flow_steer_promisc_add(mdev->dev,
791 MLX4_FS_PROMISC_UPLINK);
793 en_err(priv, "Failed enabling promiscuous mode\n");
794 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
797 case MLX4_STEERING_MODE_B0:
798 err = mlx4_unicast_promisc_add(mdev->dev,
802 en_err(priv, "Failed enabling unicast promiscuous mode\n");
804 /* Add the default qp number as multicast
807 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
808 err = mlx4_multicast_promisc_add(mdev->dev,
812 en_err(priv, "Failed enabling multicast promiscuous mode\n");
813 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
817 case MLX4_STEERING_MODE_A0:
818 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
823 en_err(priv, "Failed enabling promiscuous mode\n");
827 /* Disable port multicast filter (unconditionally) */
828 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
829 0, MLX4_MCAST_DISABLE);
831 en_err(priv, "Failed disabling multicast filter\n");
833 /* Disable port VLAN filter */
834 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
836 en_err(priv, "Failed disabling VLAN filter\n");
840 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
841 struct mlx4_en_dev *mdev)
845 if (netif_msg_rx_status(priv))
846 en_warn(priv, "Leaving promiscuous mode\n");
847 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
849 /* Disable promiscouos mode */
850 switch (mdev->dev->caps.steering_mode) {
851 case MLX4_STEERING_MODE_DEVICE_MANAGED:
852 err = mlx4_flow_steer_promisc_remove(mdev->dev,
854 MLX4_FS_PROMISC_UPLINK);
856 en_err(priv, "Failed disabling promiscuous mode\n");
857 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
860 case MLX4_STEERING_MODE_B0:
861 err = mlx4_unicast_promisc_remove(mdev->dev,
865 en_err(priv, "Failed disabling unicast promiscuous mode\n");
866 /* Disable Multicast promisc */
867 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
868 err = mlx4_multicast_promisc_remove(mdev->dev,
872 en_err(priv, "Failed disabling multicast promiscuous mode\n");
873 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
877 case MLX4_STEERING_MODE_A0:
878 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
882 en_err(priv, "Failed disabling promiscuous mode\n");
886 /* Enable port VLAN filter */
887 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
889 en_err(priv, "Failed enabling VLAN filter\n");
892 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
893 struct net_device *dev,
894 struct mlx4_en_dev *mdev)
896 struct mlx4_en_mc_list *mclist, *tmp;
898 u8 mc_list[16] = {0};
901 /* Enable/disable the multicast filter according to IFF_ALLMULTI */
902 if (dev->flags & IFF_ALLMULTI) {
903 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
904 0, MLX4_MCAST_DISABLE);
906 en_err(priv, "Failed disabling multicast filter\n");
908 /* Add the default qp number as multicast promisc */
909 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
910 switch (mdev->dev->caps.steering_mode) {
911 case MLX4_STEERING_MODE_DEVICE_MANAGED:
912 err = mlx4_flow_steer_promisc_add(mdev->dev,
915 MLX4_FS_PROMISC_ALL_MULTI);
918 case MLX4_STEERING_MODE_B0:
919 err = mlx4_multicast_promisc_add(mdev->dev,
924 case MLX4_STEERING_MODE_A0:
928 en_err(priv, "Failed entering multicast promisc mode\n");
929 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
932 /* Disable Multicast promisc */
933 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
934 switch (mdev->dev->caps.steering_mode) {
935 case MLX4_STEERING_MODE_DEVICE_MANAGED:
936 err = mlx4_flow_steer_promisc_remove(mdev->dev,
938 MLX4_FS_PROMISC_ALL_MULTI);
941 case MLX4_STEERING_MODE_B0:
942 err = mlx4_multicast_promisc_remove(mdev->dev,
947 case MLX4_STEERING_MODE_A0:
951 en_err(priv, "Failed disabling multicast promiscuous mode\n");
952 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
955 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
956 0, MLX4_MCAST_DISABLE);
958 en_err(priv, "Failed disabling multicast filter\n");
960 /* Flush mcast filter and init it with broadcast address */
961 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
962 1, MLX4_MCAST_CONFIG);
964 /* Update multicast list - we cache all addresses so they won't
965 * change while HW is updated holding the command semaphor */
966 netif_addr_lock_bh(dev);
967 mlx4_en_cache_mclist(dev);
968 netif_addr_unlock_bh(dev);
969 list_for_each_entry(mclist, &priv->mc_list, list) {
970 mcast_addr = mlx4_en_mac_to_u64(mclist->addr);
971 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
972 mcast_addr, 0, MLX4_MCAST_CONFIG);
974 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
975 0, MLX4_MCAST_ENABLE);
977 en_err(priv, "Failed enabling multicast filter\n");
979 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
980 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
981 if (mclist->action == MCLIST_REM) {
982 /* detach this address and delete from list */
983 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
984 mc_list[5] = priv->port;
985 err = mlx4_multicast_detach(mdev->dev,
986 &priv->rss_map.indir_qp,
991 en_err(priv, "Fail to detach multicast address\n");
993 /* remove from list */
994 list_del(&mclist->list);
996 } else if (mclist->action == MCLIST_ADD) {
997 /* attach the address */
998 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
999 /* needed for B0 steering support */
1000 mc_list[5] = priv->port;
1001 err = mlx4_multicast_attach(mdev->dev,
1002 &priv->rss_map.indir_qp,
1008 en_err(priv, "Fail to attach multicast address\n");
1015 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1016 struct net_device *dev,
1017 struct mlx4_en_dev *mdev)
1019 struct netdev_hw_addr *ha;
1020 struct mlx4_mac_entry *entry;
1021 struct hlist_node *tmp;
1025 struct hlist_head *bucket;
1030 /* Note that we do not need to protect our mac_hash traversal with rcu,
1031 * since all modification code is protected by mdev->state_lock
1034 /* find what to remove */
1035 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1036 bucket = &priv->mac_hash[i];
1037 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1039 netdev_for_each_uc_addr(ha, dev) {
1040 if (ether_addr_equal_64bits(entry->mac,
1047 /* MAC address of the port is not in uc list */
1048 if (ether_addr_equal_64bits(entry->mac, dev->dev_addr))
1052 mac = mlx4_en_mac_to_u64(entry->mac);
1053 mlx4_en_uc_steer_release(priv, entry->mac,
1056 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1058 hlist_del_rcu(&entry->hlist);
1059 kfree_rcu(entry, rcu);
1060 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1061 entry->mac, priv->port);
1067 /* if we didn't remove anything, there is no use in trying to add
1068 * again once we are in a forced promisc mode state
1070 if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1073 prev_flags = priv->flags;
1074 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1076 /* find what to add */
1077 netdev_for_each_uc_addr(ha, dev) {
1079 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1080 hlist_for_each_entry(entry, bucket, hlist) {
1081 if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1088 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1090 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1091 ha->addr, priv->port);
1092 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1095 mac = mlx4_en_mac_to_u64(ha->addr);
1096 memcpy(entry->mac, ha->addr, ETH_ALEN);
1097 err = mlx4_register_mac(mdev->dev, priv->port, mac);
1099 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1100 ha->addr, priv->port, err);
1102 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1105 err = mlx4_en_uc_steer_add(priv, ha->addr,
1109 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1110 ha->addr, priv->port, err);
1111 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1113 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1116 unsigned int mac_hash;
1117 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1118 ha->addr, priv->port);
1119 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1120 bucket = &priv->mac_hash[mac_hash];
1121 hlist_add_head_rcu(&entry->hlist, bucket);
1126 if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1127 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1129 } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1130 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1135 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1137 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1139 struct mlx4_en_dev *mdev = priv->mdev;
1140 struct net_device *dev = priv->dev;
1142 mutex_lock(&mdev->state_lock);
1143 if (!mdev->device_up) {
1144 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1147 if (!priv->port_up) {
1148 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1152 if (!netif_carrier_ok(dev)) {
1153 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1154 if (priv->port_state.link_state) {
1155 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1156 netif_carrier_on(dev);
1157 en_dbg(LINK, priv, "Link Up\n");
1162 if (dev->priv_flags & IFF_UNICAST_FLT)
1163 mlx4_en_do_uc_filter(priv, dev, mdev);
1165 /* Promsicuous mode: disable all filters */
1166 if ((dev->flags & IFF_PROMISC) ||
1167 (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1168 mlx4_en_set_promisc_mode(priv, mdev);
1172 /* Not in promiscuous mode */
1173 if (priv->flags & MLX4_EN_FLAG_PROMISC)
1174 mlx4_en_clear_promisc_mode(priv, mdev);
1176 mlx4_en_do_multicast(priv, dev, mdev);
1178 mutex_unlock(&mdev->state_lock);
1181 #ifdef CONFIG_NET_POLL_CONTROLLER
1182 static void mlx4_en_netpoll(struct net_device *dev)
1184 struct mlx4_en_priv *priv = netdev_priv(dev);
1185 struct mlx4_en_cq *cq;
1186 unsigned long flags;
1189 for (i = 0; i < priv->rx_ring_num; i++) {
1190 cq = &priv->rx_cq[i];
1191 spin_lock_irqsave(&cq->lock, flags);
1192 napi_synchronize(&cq->napi);
1193 mlx4_en_process_rx_cq(dev, cq, 0);
1194 spin_unlock_irqrestore(&cq->lock, flags);
1199 static void mlx4_en_tx_timeout(struct net_device *dev)
1201 struct mlx4_en_priv *priv = netdev_priv(dev);
1202 struct mlx4_en_dev *mdev = priv->mdev;
1204 if (netif_msg_timer(priv))
1205 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1207 priv->port_stats.tx_timeout++;
1208 en_dbg(DRV, priv, "Scheduling watchdog\n");
1209 queue_work(mdev->workqueue, &priv->watchdog_task);
1213 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
1215 struct mlx4_en_priv *priv = netdev_priv(dev);
1217 spin_lock_bh(&priv->stats_lock);
1218 memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
1219 spin_unlock_bh(&priv->stats_lock);
1221 return &priv->ret_stats;
1224 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1226 struct mlx4_en_cq *cq;
1229 /* If we haven't received a specific coalescing setting
1230 * (module param), we set the moderation parameters as follows:
1231 * - moder_cnt is set to the number of mtu sized packets to
1232 * satisfy our coalescing target.
1233 * - moder_time is set to a fixed value.
1235 priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1236 priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1237 priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1238 priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1239 en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1240 priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1242 /* Setup cq moderation params */
1243 for (i = 0; i < priv->rx_ring_num; i++) {
1244 cq = &priv->rx_cq[i];
1245 cq->moder_cnt = priv->rx_frames;
1246 cq->moder_time = priv->rx_usecs;
1247 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1248 priv->last_moder_packets[i] = 0;
1249 priv->last_moder_bytes[i] = 0;
1252 for (i = 0; i < priv->tx_ring_num; i++) {
1253 cq = &priv->tx_cq[i];
1254 cq->moder_cnt = priv->tx_frames;
1255 cq->moder_time = priv->tx_usecs;
1258 /* Reset auto-moderation params */
1259 priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1260 priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1261 priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1262 priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1263 priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1264 priv->adaptive_rx_coal = 1;
1265 priv->last_moder_jiffies = 0;
1266 priv->last_moder_tx_packets = 0;
1269 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1271 unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1272 struct mlx4_en_cq *cq;
1273 unsigned long packets;
1275 unsigned long avg_pkt_size;
1276 unsigned long rx_packets;
1277 unsigned long rx_bytes;
1278 unsigned long rx_pkt_diff;
1282 if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1285 for (ring = 0; ring < priv->rx_ring_num; ring++) {
1286 spin_lock_bh(&priv->stats_lock);
1287 rx_packets = priv->rx_ring[ring].packets;
1288 rx_bytes = priv->rx_ring[ring].bytes;
1289 spin_unlock_bh(&priv->stats_lock);
1291 rx_pkt_diff = ((unsigned long) (rx_packets -
1292 priv->last_moder_packets[ring]));
1293 packets = rx_pkt_diff;
1294 rate = packets * HZ / period;
1295 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1296 priv->last_moder_bytes[ring])) / packets : 0;
1298 /* Apply auto-moderation only when packet rate
1299 * exceeds a rate that it matters */
1300 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1301 avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1302 if (rate < priv->pkt_rate_low)
1303 moder_time = priv->rx_usecs_low;
1304 else if (rate > priv->pkt_rate_high)
1305 moder_time = priv->rx_usecs_high;
1307 moder_time = (rate - priv->pkt_rate_low) *
1308 (priv->rx_usecs_high - priv->rx_usecs_low) /
1309 (priv->pkt_rate_high - priv->pkt_rate_low) +
1312 moder_time = priv->rx_usecs_low;
1315 if (moder_time != priv->last_moder_time[ring]) {
1316 priv->last_moder_time[ring] = moder_time;
1317 cq = &priv->rx_cq[ring];
1318 cq->moder_time = moder_time;
1319 err = mlx4_en_set_cq_moder(priv, cq);
1321 en_err(priv, "Failed modifying moderation for cq:%d\n",
1324 priv->last_moder_packets[ring] = rx_packets;
1325 priv->last_moder_bytes[ring] = rx_bytes;
1328 priv->last_moder_jiffies = jiffies;
1331 static void mlx4_en_do_get_stats(struct work_struct *work)
1333 struct delayed_work *delay = to_delayed_work(work);
1334 struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1336 struct mlx4_en_dev *mdev = priv->mdev;
1339 mutex_lock(&mdev->state_lock);
1340 if (mdev->device_up) {
1341 err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1343 en_dbg(HW, priv, "Could not update stats\n");
1346 mlx4_en_auto_moderation(priv);
1348 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1350 if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1351 queue_work(mdev->workqueue, &priv->mac_task);
1352 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1354 mutex_unlock(&mdev->state_lock);
1357 static void mlx4_en_linkstate(struct work_struct *work)
1359 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1361 struct mlx4_en_dev *mdev = priv->mdev;
1362 int linkstate = priv->link_state;
1364 mutex_lock(&mdev->state_lock);
1365 /* If observable port state changed set carrier state and
1366 * report to system log */
1367 if (priv->last_link_state != linkstate) {
1368 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1369 en_info(priv, "Link Down\n");
1370 netif_carrier_off(priv->dev);
1372 en_info(priv, "Link Up\n");
1373 netif_carrier_on(priv->dev);
1376 priv->last_link_state = linkstate;
1377 mutex_unlock(&mdev->state_lock);
1381 int mlx4_en_start_port(struct net_device *dev)
1383 struct mlx4_en_priv *priv = netdev_priv(dev);
1384 struct mlx4_en_dev *mdev = priv->mdev;
1385 struct mlx4_en_cq *cq;
1386 struct mlx4_en_tx_ring *tx_ring;
1392 u8 mc_list[16] = {0};
1394 if (priv->port_up) {
1395 en_dbg(DRV, priv, "start port called while port already up\n");
1399 INIT_LIST_HEAD(&priv->mc_list);
1400 INIT_LIST_HEAD(&priv->curr_list);
1401 INIT_LIST_HEAD(&priv->ethtool_list);
1402 memset(&priv->ethtool_rules[0], 0,
1403 sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1405 /* Calculate Rx buf size */
1406 dev->mtu = min(dev->mtu, priv->max_mtu);
1407 mlx4_en_calc_rx_buf(dev);
1408 en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1410 /* Configure rx cq's and rings */
1411 err = mlx4_en_activate_rx_rings(priv);
1413 en_err(priv, "Failed to activate RX rings\n");
1416 for (i = 0; i < priv->rx_ring_num; i++) {
1417 cq = &priv->rx_cq[i];
1419 err = mlx4_en_activate_cq(priv, cq, i);
1421 en_err(priv, "Failed activating Rx CQ\n");
1424 for (j = 0; j < cq->size; j++)
1425 cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1426 err = mlx4_en_set_cq_moder(priv, cq);
1428 en_err(priv, "Failed setting cq moderation parameters");
1429 mlx4_en_deactivate_cq(priv, cq);
1432 mlx4_en_arm_cq(priv, cq);
1433 priv->rx_ring[i].cqn = cq->mcq.cqn;
1438 en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1439 err = mlx4_en_get_qp(priv);
1441 en_err(priv, "Failed getting eth qp\n");
1444 mdev->mac_removed[priv->port] = 0;
1446 err = mlx4_en_config_rss_steer(priv);
1448 en_err(priv, "Failed configuring rss steering\n");
1452 err = mlx4_en_create_drop_qp(priv);
1456 /* Configure tx cq's and rings */
1457 for (i = 0; i < priv->tx_ring_num; i++) {
1459 cq = &priv->tx_cq[i];
1460 err = mlx4_en_activate_cq(priv, cq, i);
1462 en_err(priv, "Failed allocating Tx CQ\n");
1465 err = mlx4_en_set_cq_moder(priv, cq);
1467 en_err(priv, "Failed setting cq moderation parameters");
1468 mlx4_en_deactivate_cq(priv, cq);
1471 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1472 cq->buf->wqe_index = cpu_to_be16(0xffff);
1474 /* Configure ring */
1475 tx_ring = &priv->tx_ring[i];
1476 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1477 i / priv->num_tx_rings_p_up);
1479 en_err(priv, "Failed allocating Tx ring\n");
1480 mlx4_en_deactivate_cq(priv, cq);
1483 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1485 /* Arm CQ for TX completions */
1486 mlx4_en_arm_cq(priv, cq);
1488 /* Set initial ownership of all Tx TXBBs to SW (1) */
1489 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1490 *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1494 /* Configure port */
1495 err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1496 priv->rx_skb_size + ETH_FCS_LEN,
1497 priv->prof->tx_pause,
1499 priv->prof->rx_pause,
1500 priv->prof->rx_ppp);
1502 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1506 /* Set default qp number */
1507 err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1509 en_err(priv, "Failed setting default qp numbers\n");
1514 en_dbg(HW, priv, "Initializing port\n");
1515 err = mlx4_INIT_PORT(mdev->dev, priv->port);
1517 en_err(priv, "Failed Initializing port\n");
1521 /* Attach rx QP to bradcast address */
1522 memset(&mc_list[10], 0xff, ETH_ALEN);
1523 mc_list[5] = priv->port; /* needed for B0 steering support */
1524 if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1525 priv->port, 0, MLX4_PROT_ETH,
1526 &priv->broadcast_id))
1527 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1529 /* Must redo promiscuous mode setup. */
1530 priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1532 /* Schedule multicast task to populate multicast list */
1533 queue_work(mdev->workqueue, &priv->rx_mode_task);
1535 mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
1537 priv->port_up = true;
1538 netif_tx_start_all_queues(dev);
1539 netif_device_attach(dev);
1544 while (tx_index--) {
1545 mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[tx_index]);
1546 mlx4_en_deactivate_cq(priv, &priv->tx_cq[tx_index]);
1548 mlx4_en_destroy_drop_qp(priv);
1550 mlx4_en_release_rss_steer(priv);
1552 mlx4_en_put_qp(priv);
1555 mlx4_en_deactivate_cq(priv, &priv->rx_cq[rx_index]);
1556 for (i = 0; i < priv->rx_ring_num; i++)
1557 mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
1559 return err; /* need to close devices */
1563 void mlx4_en_stop_port(struct net_device *dev, int detach)
1565 struct mlx4_en_priv *priv = netdev_priv(dev);
1566 struct mlx4_en_dev *mdev = priv->mdev;
1567 struct mlx4_en_mc_list *mclist, *tmp;
1568 struct ethtool_flow_id *flow, *tmp_flow;
1570 u8 mc_list[16] = {0};
1572 if (!priv->port_up) {
1573 en_dbg(DRV, priv, "stop port called while port already down\n");
1577 /* Synchronize with tx routine */
1578 netif_tx_lock_bh(dev);
1580 netif_device_detach(dev);
1581 netif_tx_stop_all_queues(dev);
1582 netif_tx_unlock_bh(dev);
1584 netif_tx_disable(dev);
1586 /* Set port as not active */
1587 priv->port_up = false;
1589 /* Promsicuous mode */
1590 if (mdev->dev->caps.steering_mode ==
1591 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1592 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1593 MLX4_EN_FLAG_MC_PROMISC);
1594 mlx4_flow_steer_promisc_remove(mdev->dev,
1596 MLX4_FS_PROMISC_UPLINK);
1597 mlx4_flow_steer_promisc_remove(mdev->dev,
1599 MLX4_FS_PROMISC_ALL_MULTI);
1600 } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1601 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1603 /* Disable promiscouos mode */
1604 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1607 /* Disable Multicast promisc */
1608 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1609 mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1611 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1615 /* Detach All multicasts */
1616 memset(&mc_list[10], 0xff, ETH_ALEN);
1617 mc_list[5] = priv->port; /* needed for B0 steering support */
1618 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1619 MLX4_PROT_ETH, priv->broadcast_id);
1620 list_for_each_entry(mclist, &priv->curr_list, list) {
1621 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1622 mc_list[5] = priv->port;
1623 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1624 mc_list, MLX4_PROT_ETH, mclist->reg_id);
1626 mlx4_en_clear_list(dev);
1627 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1628 list_del(&mclist->list);
1632 /* Flush multicast filter */
1633 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1635 mlx4_en_destroy_drop_qp(priv);
1638 for (i = 0; i < priv->tx_ring_num; i++) {
1639 mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[i]);
1640 mlx4_en_deactivate_cq(priv, &priv->tx_cq[i]);
1644 for (i = 0; i < priv->tx_ring_num; i++)
1645 mlx4_en_free_tx_buf(dev, &priv->tx_ring[i]);
1648 mlx4_en_release_rss_steer(priv);
1650 /* Unregister Mac address for the port */
1651 mlx4_en_put_qp(priv);
1652 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAGS2_REASSIGN_MAC_EN))
1653 mdev->mac_removed[priv->port] = 1;
1655 /* Remove flow steering rules for the port*/
1656 if (mdev->dev->caps.steering_mode ==
1657 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1659 list_for_each_entry_safe(flow, tmp_flow,
1660 &priv->ethtool_list, list) {
1661 mlx4_flow_detach(mdev->dev, flow->id);
1662 list_del(&flow->list);
1667 for (i = 0; i < priv->rx_ring_num; i++) {
1668 mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
1669 while (test_bit(NAPI_STATE_SCHED, &priv->rx_cq[i].napi.state))
1671 mlx4_en_deactivate_cq(priv, &priv->rx_cq[i]);
1675 mlx4_CLOSE_PORT(mdev->dev, priv->port);
1678 static void mlx4_en_restart(struct work_struct *work)
1680 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1682 struct mlx4_en_dev *mdev = priv->mdev;
1683 struct net_device *dev = priv->dev;
1685 en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1687 mutex_lock(&mdev->state_lock);
1688 if (priv->port_up) {
1689 mlx4_en_stop_port(dev, 1);
1690 if (mlx4_en_start_port(dev))
1691 en_err(priv, "Failed restarting port %d\n", priv->port);
1693 mutex_unlock(&mdev->state_lock);
1696 static void mlx4_en_clear_stats(struct net_device *dev)
1698 struct mlx4_en_priv *priv = netdev_priv(dev);
1699 struct mlx4_en_dev *mdev = priv->mdev;
1702 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1703 en_dbg(HW, priv, "Failed dumping statistics\n");
1705 memset(&priv->stats, 0, sizeof(priv->stats));
1706 memset(&priv->pstats, 0, sizeof(priv->pstats));
1707 memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1708 memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1710 for (i = 0; i < priv->tx_ring_num; i++) {
1711 priv->tx_ring[i].bytes = 0;
1712 priv->tx_ring[i].packets = 0;
1713 priv->tx_ring[i].tx_csum = 0;
1715 for (i = 0; i < priv->rx_ring_num; i++) {
1716 priv->rx_ring[i].bytes = 0;
1717 priv->rx_ring[i].packets = 0;
1718 priv->rx_ring[i].csum_ok = 0;
1719 priv->rx_ring[i].csum_none = 0;
1723 static int mlx4_en_open(struct net_device *dev)
1725 struct mlx4_en_priv *priv = netdev_priv(dev);
1726 struct mlx4_en_dev *mdev = priv->mdev;
1729 mutex_lock(&mdev->state_lock);
1731 if (!mdev->device_up) {
1732 en_err(priv, "Cannot open - device down/disabled\n");
1737 /* Reset HW statistics and SW counters */
1738 mlx4_en_clear_stats(dev);
1740 err = mlx4_en_start_port(dev);
1742 en_err(priv, "Failed starting port:%d\n", priv->port);
1745 mutex_unlock(&mdev->state_lock);
1750 static int mlx4_en_close(struct net_device *dev)
1752 struct mlx4_en_priv *priv = netdev_priv(dev);
1753 struct mlx4_en_dev *mdev = priv->mdev;
1755 en_dbg(IFDOWN, priv, "Close port called\n");
1757 mutex_lock(&mdev->state_lock);
1759 mlx4_en_stop_port(dev, 0);
1760 netif_carrier_off(dev);
1762 mutex_unlock(&mdev->state_lock);
1766 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1770 #ifdef CONFIG_RFS_ACCEL
1771 free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1772 priv->dev->rx_cpu_rmap = NULL;
1775 for (i = 0; i < priv->tx_ring_num; i++) {
1776 if (priv->tx_ring[i].tx_info)
1777 mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1778 if (priv->tx_cq[i].buf)
1779 mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1782 for (i = 0; i < priv->rx_ring_num; i++) {
1783 if (priv->rx_ring[i].rx_info)
1784 mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1785 priv->prof->rx_ring_size, priv->stride);
1786 if (priv->rx_cq[i].buf)
1787 mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1790 if (priv->base_tx_qpn) {
1791 mlx4_qp_release_range(priv->mdev->dev, priv->base_tx_qpn, priv->tx_ring_num);
1792 priv->base_tx_qpn = 0;
1796 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1798 struct mlx4_en_port_profile *prof = priv->prof;
1802 err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &priv->base_tx_qpn);
1804 en_err(priv, "failed reserving range for TX rings\n");
1808 /* Create tx Rings */
1809 for (i = 0; i < priv->tx_ring_num; i++) {
1810 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1811 prof->tx_ring_size, i, TX))
1814 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i], priv->base_tx_qpn + i,
1815 prof->tx_ring_size, TXBB_SIZE))
1819 /* Create rx Rings */
1820 for (i = 0; i < priv->rx_ring_num; i++) {
1821 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
1822 prof->rx_ring_size, i, RX))
1825 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
1826 prof->rx_ring_size, priv->stride))
1830 #ifdef CONFIG_RFS_ACCEL
1831 priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->mdev->dev->caps.comp_pool);
1832 if (!priv->dev->rx_cpu_rmap)
1839 en_err(priv, "Failed to allocate NIC resources\n");
1844 void mlx4_en_destroy_netdev(struct net_device *dev)
1846 struct mlx4_en_priv *priv = netdev_priv(dev);
1847 struct mlx4_en_dev *mdev = priv->mdev;
1849 en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
1851 /* Unregister device - this will close the port if it was up */
1852 if (priv->registered)
1853 unregister_netdev(dev);
1855 if (priv->allocated)
1856 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
1858 cancel_delayed_work(&priv->stats_task);
1859 /* flush any pending task for this netdev */
1860 flush_workqueue(mdev->workqueue);
1862 /* Detach the netdev so tasks would not attempt to access it */
1863 mutex_lock(&mdev->state_lock);
1864 mdev->pndev[priv->port] = NULL;
1865 mutex_unlock(&mdev->state_lock);
1867 mlx4_en_free_resources(priv);
1869 kfree(priv->tx_ring);
1875 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
1877 struct mlx4_en_priv *priv = netdev_priv(dev);
1878 struct mlx4_en_dev *mdev = priv->mdev;
1881 en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
1884 if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
1885 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
1890 if (netif_running(dev)) {
1891 mutex_lock(&mdev->state_lock);
1892 if (!mdev->device_up) {
1893 /* NIC is probably restarting - let watchdog task reset
1895 en_dbg(DRV, priv, "Change MTU called with card down!?\n");
1897 mlx4_en_stop_port(dev, 1);
1898 err = mlx4_en_start_port(dev);
1900 en_err(priv, "Failed restarting port:%d\n",
1902 queue_work(mdev->workqueue, &priv->watchdog_task);
1905 mutex_unlock(&mdev->state_lock);
1910 static int mlx4_en_set_features(struct net_device *netdev,
1911 netdev_features_t features)
1913 struct mlx4_en_priv *priv = netdev_priv(netdev);
1915 if (features & NETIF_F_LOOPBACK)
1916 priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
1919 cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
1921 mlx4_en_update_loopback_state(netdev, features);
1927 static int mlx4_en_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1928 struct net_device *dev,
1929 const unsigned char *addr, u16 flags)
1931 struct mlx4_en_priv *priv = netdev_priv(dev);
1932 struct mlx4_dev *mdev = priv->mdev->dev;
1935 if (!mlx4_is_mfunc(mdev))
1938 /* Hardware does not support aging addresses, allow only
1939 * permanent addresses if ndm_state is given
1941 if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
1942 en_info(priv, "Add FDB only supports static addresses\n");
1946 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr))
1947 err = dev_uc_add_excl(dev, addr);
1948 else if (is_multicast_ether_addr(addr))
1949 err = dev_mc_add_excl(dev, addr);
1953 /* Only return duplicate errors if NLM_F_EXCL is set */
1954 if (err == -EEXIST && !(flags & NLM_F_EXCL))
1960 static int mlx4_en_fdb_del(struct ndmsg *ndm,
1961 struct nlattr *tb[],
1962 struct net_device *dev,
1963 const unsigned char *addr)
1965 struct mlx4_en_priv *priv = netdev_priv(dev);
1966 struct mlx4_dev *mdev = priv->mdev->dev;
1969 if (!mlx4_is_mfunc(mdev))
1972 if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
1973 en_info(priv, "Del FDB only supports static addresses\n");
1977 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr))
1978 err = dev_uc_del(dev, addr);
1979 else if (is_multicast_ether_addr(addr))
1980 err = dev_mc_del(dev, addr);
1987 static int mlx4_en_fdb_dump(struct sk_buff *skb,
1988 struct netlink_callback *cb,
1989 struct net_device *dev, int idx)
1991 struct mlx4_en_priv *priv = netdev_priv(dev);
1992 struct mlx4_dev *mdev = priv->mdev->dev;
1994 if (mlx4_is_mfunc(mdev))
1995 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
2000 static const struct net_device_ops mlx4_netdev_ops = {
2001 .ndo_open = mlx4_en_open,
2002 .ndo_stop = mlx4_en_close,
2003 .ndo_start_xmit = mlx4_en_xmit,
2004 .ndo_select_queue = mlx4_en_select_queue,
2005 .ndo_get_stats = mlx4_en_get_stats,
2006 .ndo_set_rx_mode = mlx4_en_set_rx_mode,
2007 .ndo_set_mac_address = mlx4_en_set_mac,
2008 .ndo_validate_addr = eth_validate_addr,
2009 .ndo_change_mtu = mlx4_en_change_mtu,
2010 .ndo_tx_timeout = mlx4_en_tx_timeout,
2011 .ndo_vlan_rx_add_vid = mlx4_en_vlan_rx_add_vid,
2012 .ndo_vlan_rx_kill_vid = mlx4_en_vlan_rx_kill_vid,
2013 #ifdef CONFIG_NET_POLL_CONTROLLER
2014 .ndo_poll_controller = mlx4_en_netpoll,
2016 .ndo_set_features = mlx4_en_set_features,
2017 .ndo_setup_tc = mlx4_en_setup_tc,
2018 #ifdef CONFIG_RFS_ACCEL
2019 .ndo_rx_flow_steer = mlx4_en_filter_rfs,
2021 .ndo_fdb_add = mlx4_en_fdb_add,
2022 .ndo_fdb_del = mlx4_en_fdb_del,
2023 .ndo_fdb_dump = mlx4_en_fdb_dump,
2026 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2027 struct mlx4_en_port_profile *prof)
2029 struct net_device *dev;
2030 struct mlx4_en_priv *priv;
2034 dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2035 MAX_TX_RINGS, MAX_RX_RINGS);
2039 netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2040 netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2042 SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
2043 dev->dev_id = port - 1;
2046 * Initialize driver private data
2049 priv = netdev_priv(dev);
2050 memset(priv, 0, sizeof(struct mlx4_en_priv));
2053 priv->ddev = &mdev->pdev->dev;
2056 priv->port_up = false;
2057 priv->flags = prof->flags;
2058 priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2059 MLX4_WQE_CTRL_SOLICITED);
2060 priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2061 priv->tx_ring_num = prof->tx_ring_num;
2063 priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring) * MAX_TX_RINGS,
2065 if (!priv->tx_ring) {
2069 priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq) * MAX_TX_RINGS,
2075 priv->rx_ring_num = prof->rx_ring_num;
2076 priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2077 priv->mac_index = -1;
2078 priv->msg_enable = MLX4_EN_MSG_LEVEL;
2079 spin_lock_init(&priv->stats_lock);
2080 INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2081 INIT_WORK(&priv->mac_task, mlx4_en_do_set_mac);
2082 INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2083 INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2084 INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2085 #ifdef CONFIG_MLX4_EN_DCB
2086 if (!mlx4_is_slave(priv->mdev->dev))
2087 dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2090 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2091 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2093 /* Query for default mac and max mtu */
2094 priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2096 /* Set default MAC */
2097 dev->addr_len = ETH_ALEN;
2098 mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2099 if (!is_valid_ether_addr(dev->dev_addr)) {
2100 en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2101 priv->port, dev->dev_addr);
2106 memcpy(priv->prev_mac, dev->dev_addr, sizeof(priv->prev_mac));
2108 priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2109 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2110 err = mlx4_en_alloc_resources(priv);
2114 #ifdef CONFIG_RFS_ACCEL
2115 INIT_LIST_HEAD(&priv->filters);
2116 spin_lock_init(&priv->filters_lock);
2119 /* Allocate page for receive rings */
2120 err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2121 MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2123 en_err(priv, "Failed to allocate page for rx qps\n");
2126 priv->allocated = 1;
2129 * Initialize netdev entry points
2131 dev->netdev_ops = &mlx4_netdev_ops;
2132 dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2133 netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2134 netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2136 SET_ETHTOOL_OPS(dev, &mlx4_en_ethtool_ops);
2139 * Set driver features
2141 dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2142 if (mdev->LSO_support)
2143 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2145 dev->vlan_features = dev->hw_features;
2147 dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2148 dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2149 NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
2150 NETIF_F_HW_VLAN_FILTER;
2151 dev->hw_features |= NETIF_F_LOOPBACK;
2153 if (mdev->dev->caps.steering_mode ==
2154 MLX4_STEERING_MODE_DEVICE_MANAGED)
2155 dev->hw_features |= NETIF_F_NTUPLE;
2157 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
2158 dev->priv_flags |= IFF_UNICAST_FLT;
2160 mdev->pndev[port] = dev;
2162 netif_carrier_off(dev);
2163 err = register_netdev(dev);
2165 en_err(priv, "Netdev registration failed for port %d\n", port);
2168 priv->registered = 1;
2170 en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2171 en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2173 mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
2175 /* Configure port */
2176 mlx4_en_calc_rx_buf(dev);
2177 err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2178 priv->rx_skb_size + ETH_FCS_LEN,
2179 prof->tx_pause, prof->tx_ppp,
2180 prof->rx_pause, prof->rx_ppp);
2182 en_err(priv, "Failed setting port general configurations "
2183 "for port %d, with error %d\n", priv->port, err);
2188 en_warn(priv, "Initializing port\n");
2189 err = mlx4_INIT_PORT(mdev->dev, priv->port);
2191 en_err(priv, "Failed Initializing port\n");
2194 mlx4_en_set_default_moderation(priv);
2195 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2199 mlx4_en_destroy_netdev(dev);