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>
41 #include <net/busy_poll.h>
42 #include <net/vxlan.h>
43 #include <net/devlink.h>
45 #include <linux/mlx4/driver.h>
46 #include <linux/mlx4/device.h>
47 #include <linux/mlx4/cmd.h>
48 #include <linux/mlx4/cq.h>
53 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
55 struct mlx4_en_priv *priv = netdev_priv(dev);
57 unsigned int offset = 0;
59 if (up && up != MLX4_EN_NUM_UP)
62 netdev_set_num_tc(dev, up);
64 /* Partition Tx queues evenly amongst UP's */
65 for (i = 0; i < up; i++) {
66 netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
67 offset += priv->num_tx_rings_p_up;
73 static int __mlx4_en_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
74 struct tc_to_netdev *tc)
76 if (tc->type != TC_SETUP_MQPRIO)
79 return mlx4_en_setup_tc(dev, tc->tc);
82 #ifdef CONFIG_RFS_ACCEL
84 struct mlx4_en_filter {
85 struct list_head next;
86 struct work_struct work;
95 struct mlx4_en_priv *priv;
96 u32 flow_id; /* RFS infrastructure id */
97 int id; /* mlx4_en driver id */
98 u64 reg_id; /* Flow steering API id */
99 u8 activated; /* Used to prevent expiry before filter
102 struct hlist_node filter_chain;
105 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
107 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
111 return MLX4_NET_TRANS_RULE_ID_UDP;
113 return MLX4_NET_TRANS_RULE_ID_TCP;
115 return MLX4_NET_TRANS_RULE_NUM;
119 static void mlx4_en_filter_work(struct work_struct *work)
121 struct mlx4_en_filter *filter = container_of(work,
122 struct mlx4_en_filter,
124 struct mlx4_en_priv *priv = filter->priv;
125 struct mlx4_spec_list spec_tcp_udp = {
126 .id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
129 .dst_port = filter->dst_port,
130 .dst_port_msk = (__force __be16)-1,
131 .src_port = filter->src_port,
132 .src_port_msk = (__force __be16)-1,
136 struct mlx4_spec_list spec_ip = {
137 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
140 .dst_ip = filter->dst_ip,
141 .dst_ip_msk = (__force __be32)-1,
142 .src_ip = filter->src_ip,
143 .src_ip_msk = (__force __be32)-1,
147 struct mlx4_spec_list spec_eth = {
148 .id = MLX4_NET_TRANS_RULE_ID_ETH,
150 struct mlx4_net_trans_rule rule = {
151 .list = LIST_HEAD_INIT(rule.list),
152 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
155 .promisc_mode = MLX4_FS_REGULAR,
157 .priority = MLX4_DOMAIN_RFS,
160 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
162 if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
163 en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
167 list_add_tail(&spec_eth.list, &rule.list);
168 list_add_tail(&spec_ip.list, &rule.list);
169 list_add_tail(&spec_tcp_udp.list, &rule.list);
171 rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
172 memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
173 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
175 filter->activated = 0;
177 if (filter->reg_id) {
178 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
179 if (rc && rc != -ENOENT)
180 en_err(priv, "Error detaching flow. rc = %d\n", rc);
183 rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
185 en_err(priv, "Error attaching flow. err = %d\n", rc);
188 mlx4_en_filter_rfs_expire(priv);
190 filter->activated = 1;
193 static inline struct hlist_head *
194 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
195 __be16 src_port, __be16 dst_port)
200 l = (__force unsigned long)src_port |
201 ((__force unsigned long)dst_port << 2);
202 l ^= (__force unsigned long)(src_ip ^ dst_ip);
204 bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
206 return &priv->filter_hash[bucket_idx];
209 static struct mlx4_en_filter *
210 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
211 __be32 dst_ip, u8 ip_proto, __be16 src_port,
212 __be16 dst_port, u32 flow_id)
214 struct mlx4_en_filter *filter = NULL;
216 filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
221 filter->rxq_index = rxq_index;
222 INIT_WORK(&filter->work, mlx4_en_filter_work);
224 filter->src_ip = src_ip;
225 filter->dst_ip = dst_ip;
226 filter->ip_proto = ip_proto;
227 filter->src_port = src_port;
228 filter->dst_port = dst_port;
230 filter->flow_id = flow_id;
232 filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
234 list_add_tail(&filter->next, &priv->filters);
235 hlist_add_head(&filter->filter_chain,
236 filter_hash_bucket(priv, src_ip, dst_ip, src_port,
242 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
244 struct mlx4_en_priv *priv = filter->priv;
247 list_del(&filter->next);
249 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
250 if (rc && rc != -ENOENT)
251 en_err(priv, "Error detaching flow. rc = %d\n", rc);
256 static inline struct mlx4_en_filter *
257 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
258 u8 ip_proto, __be16 src_port, __be16 dst_port)
260 struct mlx4_en_filter *filter;
261 struct mlx4_en_filter *ret = NULL;
263 hlist_for_each_entry(filter,
264 filter_hash_bucket(priv, src_ip, dst_ip,
267 if (filter->src_ip == src_ip &&
268 filter->dst_ip == dst_ip &&
269 filter->ip_proto == ip_proto &&
270 filter->src_port == src_port &&
271 filter->dst_port == dst_port) {
281 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
282 u16 rxq_index, u32 flow_id)
284 struct mlx4_en_priv *priv = netdev_priv(net_dev);
285 struct mlx4_en_filter *filter;
286 const struct iphdr *ip;
293 int nhoff = skb_network_offset(skb);
296 if (skb->protocol != htons(ETH_P_IP))
297 return -EPROTONOSUPPORT;
299 ip = (const struct iphdr *)(skb->data + nhoff);
300 if (ip_is_fragment(ip))
301 return -EPROTONOSUPPORT;
303 if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
304 return -EPROTONOSUPPORT;
305 ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
307 ip_proto = ip->protocol;
313 spin_lock_bh(&priv->filters_lock);
314 filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
317 if (filter->rxq_index == rxq_index)
320 filter->rxq_index = rxq_index;
322 filter = mlx4_en_filter_alloc(priv, rxq_index,
323 src_ip, dst_ip, ip_proto,
324 src_port, dst_port, flow_id);
331 queue_work(priv->mdev->workqueue, &filter->work);
336 spin_unlock_bh(&priv->filters_lock);
341 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
343 struct mlx4_en_filter *filter, *tmp;
346 spin_lock_bh(&priv->filters_lock);
347 list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
348 list_move(&filter->next, &del_list);
349 hlist_del(&filter->filter_chain);
351 spin_unlock_bh(&priv->filters_lock);
353 list_for_each_entry_safe(filter, tmp, &del_list, next) {
354 cancel_work_sync(&filter->work);
355 mlx4_en_filter_free(filter);
359 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
361 struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
365 spin_lock_bh(&priv->filters_lock);
366 list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
367 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
370 if (filter->activated &&
371 !work_pending(&filter->work) &&
372 rps_may_expire_flow(priv->dev,
373 filter->rxq_index, filter->flow_id,
375 list_move(&filter->next, &del_list);
376 hlist_del(&filter->filter_chain);
378 last_filter = filter;
383 if (last_filter && (&last_filter->next != priv->filters.next))
384 list_move(&priv->filters, &last_filter->next);
386 spin_unlock_bh(&priv->filters_lock);
388 list_for_each_entry_safe(filter, tmp, &del_list, next)
389 mlx4_en_filter_free(filter);
393 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
394 __be16 proto, u16 vid)
396 struct mlx4_en_priv *priv = netdev_priv(dev);
397 struct mlx4_en_dev *mdev = priv->mdev;
401 en_dbg(HW, priv, "adding VLAN:%d\n", vid);
403 set_bit(vid, priv->active_vlans);
405 /* Add VID to port VLAN filter */
406 mutex_lock(&mdev->state_lock);
407 if (mdev->device_up && priv->port_up) {
408 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
410 en_err(priv, "Failed configuring VLAN filter\n");
412 if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
413 en_dbg(HW, priv, "failed adding vlan %d\n", vid);
414 mutex_unlock(&mdev->state_lock);
419 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
420 __be16 proto, u16 vid)
422 struct mlx4_en_priv *priv = netdev_priv(dev);
423 struct mlx4_en_dev *mdev = priv->mdev;
426 en_dbg(HW, priv, "Killing VID:%d\n", vid);
428 clear_bit(vid, priv->active_vlans);
430 /* Remove VID from port VLAN filter */
431 mutex_lock(&mdev->state_lock);
432 mlx4_unregister_vlan(mdev->dev, priv->port, vid);
434 if (mdev->device_up && priv->port_up) {
435 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
437 en_err(priv, "Failed configuring VLAN filter\n");
439 mutex_unlock(&mdev->state_lock);
444 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
447 for (i = ETH_ALEN - 1; i >= 0; --i) {
448 dst_mac[i] = src_mac & 0xff;
451 memset(&dst_mac[ETH_ALEN], 0, 2);
455 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
456 int qpn, u64 *reg_id)
460 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
461 priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
462 return 0; /* do nothing */
464 err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
465 MLX4_DOMAIN_NIC, reg_id);
467 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
470 en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
475 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
476 unsigned char *mac, int *qpn, u64 *reg_id)
478 struct mlx4_en_dev *mdev = priv->mdev;
479 struct mlx4_dev *dev = mdev->dev;
482 switch (dev->caps.steering_mode) {
483 case MLX4_STEERING_MODE_B0: {
488 memcpy(&gid[10], mac, ETH_ALEN);
491 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
494 case MLX4_STEERING_MODE_DEVICE_MANAGED: {
495 struct mlx4_spec_list spec_eth = { {NULL} };
496 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
498 struct mlx4_net_trans_rule rule = {
499 .queue_mode = MLX4_NET_TRANS_Q_FIFO,
502 .promisc_mode = MLX4_FS_REGULAR,
503 .priority = MLX4_DOMAIN_NIC,
506 rule.port = priv->port;
508 INIT_LIST_HEAD(&rule.list);
510 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
511 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
512 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
513 list_add_tail(&spec_eth.list, &rule.list);
515 err = mlx4_flow_attach(dev, &rule, reg_id);
522 en_warn(priv, "Failed Attaching Unicast\n");
527 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
528 unsigned char *mac, int qpn, u64 reg_id)
530 struct mlx4_en_dev *mdev = priv->mdev;
531 struct mlx4_dev *dev = mdev->dev;
533 switch (dev->caps.steering_mode) {
534 case MLX4_STEERING_MODE_B0: {
539 memcpy(&gid[10], mac, ETH_ALEN);
542 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
545 case MLX4_STEERING_MODE_DEVICE_MANAGED: {
546 mlx4_flow_detach(dev, reg_id);
550 en_err(priv, "Invalid steering mode.\n");
554 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
556 struct mlx4_en_dev *mdev = priv->mdev;
557 struct mlx4_dev *dev = mdev->dev;
560 int *qpn = &priv->base_qpn;
561 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
563 en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
564 priv->dev->dev_addr);
565 index = mlx4_register_mac(dev, priv->port, mac);
568 en_err(priv, "Failed adding MAC: %pM\n",
569 priv->dev->dev_addr);
573 if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
574 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
575 *qpn = base_qpn + index;
579 err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP);
580 en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
582 en_err(priv, "Failed to reserve qp for mac registration\n");
583 mlx4_unregister_mac(dev, priv->port, mac);
590 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
592 struct mlx4_en_dev *mdev = priv->mdev;
593 struct mlx4_dev *dev = mdev->dev;
594 int qpn = priv->base_qpn;
596 if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
597 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
598 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
599 priv->dev->dev_addr);
600 mlx4_unregister_mac(dev, priv->port, mac);
602 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
604 mlx4_qp_release_range(dev, qpn, 1);
605 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
609 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
610 unsigned char *new_mac, unsigned char *prev_mac)
612 struct mlx4_en_dev *mdev = priv->mdev;
613 struct mlx4_dev *dev = mdev->dev;
615 u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
617 if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
618 struct hlist_head *bucket;
619 unsigned int mac_hash;
620 struct mlx4_mac_entry *entry;
621 struct hlist_node *tmp;
622 u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
624 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
625 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
626 if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
627 mlx4_en_uc_steer_release(priv, entry->mac,
629 mlx4_unregister_mac(dev, priv->port,
631 hlist_del_rcu(&entry->hlist);
633 memcpy(entry->mac, new_mac, ETH_ALEN);
635 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
636 hlist_add_head_rcu(&entry->hlist,
637 &priv->mac_hash[mac_hash]);
638 mlx4_register_mac(dev, priv->port, new_mac_u64);
639 err = mlx4_en_uc_steer_add(priv, new_mac,
644 if (priv->tunnel_reg_id) {
645 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
646 priv->tunnel_reg_id = 0;
648 err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
649 &priv->tunnel_reg_id);
656 return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
659 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
660 unsigned char new_mac[ETH_ALEN + 2])
665 /* Remove old MAC and insert the new one */
666 err = mlx4_en_replace_mac(priv, priv->base_qpn,
667 new_mac, priv->current_mac);
669 en_err(priv, "Failed changing HW MAC address\n");
671 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
674 memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
679 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
681 struct mlx4_en_priv *priv = netdev_priv(dev);
682 struct mlx4_en_dev *mdev = priv->mdev;
683 struct sockaddr *saddr = addr;
684 unsigned char new_mac[ETH_ALEN + 2];
687 if (!is_valid_ether_addr(saddr->sa_data))
688 return -EADDRNOTAVAIL;
690 mutex_lock(&mdev->state_lock);
691 memcpy(new_mac, saddr->sa_data, ETH_ALEN);
692 err = mlx4_en_do_set_mac(priv, new_mac);
694 memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
695 mutex_unlock(&mdev->state_lock);
700 static void mlx4_en_clear_list(struct net_device *dev)
702 struct mlx4_en_priv *priv = netdev_priv(dev);
703 struct mlx4_en_mc_list *tmp, *mc_to_del;
705 list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
706 list_del(&mc_to_del->list);
711 static void mlx4_en_cache_mclist(struct net_device *dev)
713 struct mlx4_en_priv *priv = netdev_priv(dev);
714 struct netdev_hw_addr *ha;
715 struct mlx4_en_mc_list *tmp;
717 mlx4_en_clear_list(dev);
718 netdev_for_each_mc_addr(ha, dev) {
719 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
721 mlx4_en_clear_list(dev);
724 memcpy(tmp->addr, ha->addr, ETH_ALEN);
725 list_add_tail(&tmp->list, &priv->mc_list);
729 static void update_mclist_flags(struct mlx4_en_priv *priv,
730 struct list_head *dst,
731 struct list_head *src)
733 struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
736 /* Find all the entries that should be removed from dst,
737 * These are the entries that are not found in src
739 list_for_each_entry(dst_tmp, dst, list) {
741 list_for_each_entry(src_tmp, src, list) {
742 if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
748 dst_tmp->action = MCLIST_REM;
751 /* Add entries that exist in src but not in dst
752 * mark them as need to add
754 list_for_each_entry(src_tmp, src, list) {
756 list_for_each_entry(dst_tmp, dst, list) {
757 if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
758 dst_tmp->action = MCLIST_NONE;
764 new_mc = kmemdup(src_tmp,
765 sizeof(struct mlx4_en_mc_list),
770 new_mc->action = MCLIST_ADD;
771 list_add_tail(&new_mc->list, dst);
776 static void mlx4_en_set_rx_mode(struct net_device *dev)
778 struct mlx4_en_priv *priv = netdev_priv(dev);
783 queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
786 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
787 struct mlx4_en_dev *mdev)
791 if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
792 if (netif_msg_rx_status(priv))
793 en_warn(priv, "Entering promiscuous mode\n");
794 priv->flags |= MLX4_EN_FLAG_PROMISC;
796 /* Enable promiscouos mode */
797 switch (mdev->dev->caps.steering_mode) {
798 case MLX4_STEERING_MODE_DEVICE_MANAGED:
799 err = mlx4_flow_steer_promisc_add(mdev->dev,
802 MLX4_FS_ALL_DEFAULT);
804 en_err(priv, "Failed enabling promiscuous mode\n");
805 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
808 case MLX4_STEERING_MODE_B0:
809 err = mlx4_unicast_promisc_add(mdev->dev,
813 en_err(priv, "Failed enabling unicast promiscuous mode\n");
815 /* Add the default qp number as multicast
818 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
819 err = mlx4_multicast_promisc_add(mdev->dev,
823 en_err(priv, "Failed enabling multicast promiscuous mode\n");
824 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
828 case MLX4_STEERING_MODE_A0:
829 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
834 en_err(priv, "Failed enabling promiscuous mode\n");
838 /* Disable port multicast filter (unconditionally) */
839 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
840 0, MLX4_MCAST_DISABLE);
842 en_err(priv, "Failed disabling multicast filter\n");
846 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
847 struct mlx4_en_dev *mdev)
851 if (netif_msg_rx_status(priv))
852 en_warn(priv, "Leaving promiscuous mode\n");
853 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
855 /* Disable promiscouos mode */
856 switch (mdev->dev->caps.steering_mode) {
857 case MLX4_STEERING_MODE_DEVICE_MANAGED:
858 err = mlx4_flow_steer_promisc_remove(mdev->dev,
860 MLX4_FS_ALL_DEFAULT);
862 en_err(priv, "Failed disabling promiscuous mode\n");
863 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
866 case MLX4_STEERING_MODE_B0:
867 err = mlx4_unicast_promisc_remove(mdev->dev,
871 en_err(priv, "Failed disabling unicast promiscuous mode\n");
872 /* Disable Multicast promisc */
873 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
874 err = mlx4_multicast_promisc_remove(mdev->dev,
878 en_err(priv, "Failed disabling multicast promiscuous mode\n");
879 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
883 case MLX4_STEERING_MODE_A0:
884 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
888 en_err(priv, "Failed disabling promiscuous mode\n");
893 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
894 struct net_device *dev,
895 struct mlx4_en_dev *mdev)
897 struct mlx4_en_mc_list *mclist, *tmp;
899 u8 mc_list[16] = {0};
902 /* Enable/disable the multicast filter according to IFF_ALLMULTI */
903 if (dev->flags & IFF_ALLMULTI) {
904 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
905 0, MLX4_MCAST_DISABLE);
907 en_err(priv, "Failed disabling multicast filter\n");
909 /* Add the default qp number as multicast promisc */
910 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
911 switch (mdev->dev->caps.steering_mode) {
912 case MLX4_STEERING_MODE_DEVICE_MANAGED:
913 err = mlx4_flow_steer_promisc_add(mdev->dev,
919 case MLX4_STEERING_MODE_B0:
920 err = mlx4_multicast_promisc_add(mdev->dev,
925 case MLX4_STEERING_MODE_A0:
929 en_err(priv, "Failed entering multicast promisc mode\n");
930 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
933 /* Disable Multicast promisc */
934 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
935 switch (mdev->dev->caps.steering_mode) {
936 case MLX4_STEERING_MODE_DEVICE_MANAGED:
937 err = mlx4_flow_steer_promisc_remove(mdev->dev,
942 case MLX4_STEERING_MODE_B0:
943 err = mlx4_multicast_promisc_remove(mdev->dev,
948 case MLX4_STEERING_MODE_A0:
952 en_err(priv, "Failed disabling multicast promiscuous mode\n");
953 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
956 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
957 0, MLX4_MCAST_DISABLE);
959 en_err(priv, "Failed disabling multicast filter\n");
961 /* Flush mcast filter and init it with broadcast address */
962 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
963 1, MLX4_MCAST_CONFIG);
965 /* Update multicast list - we cache all addresses so they won't
966 * change while HW is updated holding the command semaphor */
967 netif_addr_lock_bh(dev);
968 mlx4_en_cache_mclist(dev);
969 netif_addr_unlock_bh(dev);
970 list_for_each_entry(mclist, &priv->mc_list, list) {
971 mcast_addr = mlx4_mac_to_u64(mclist->addr);
972 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
973 mcast_addr, 0, MLX4_MCAST_CONFIG);
975 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
976 0, MLX4_MCAST_ENABLE);
978 en_err(priv, "Failed enabling multicast filter\n");
980 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
981 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
982 if (mclist->action == MCLIST_REM) {
983 /* detach this address and delete from list */
984 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
985 mc_list[5] = priv->port;
986 err = mlx4_multicast_detach(mdev->dev,
987 &priv->rss_map.indir_qp,
992 en_err(priv, "Fail to detach multicast address\n");
994 if (mclist->tunnel_reg_id) {
995 err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
997 en_err(priv, "Failed to detach multicast address\n");
1000 /* remove from list */
1001 list_del(&mclist->list);
1003 } else if (mclist->action == MCLIST_ADD) {
1004 /* attach the address */
1005 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1006 /* needed for B0 steering support */
1007 mc_list[5] = priv->port;
1008 err = mlx4_multicast_attach(mdev->dev,
1009 &priv->rss_map.indir_qp,
1015 en_err(priv, "Fail to attach multicast address\n");
1017 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1018 &mclist->tunnel_reg_id);
1020 en_err(priv, "Failed to attach multicast address\n");
1026 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1027 struct net_device *dev,
1028 struct mlx4_en_dev *mdev)
1030 struct netdev_hw_addr *ha;
1031 struct mlx4_mac_entry *entry;
1032 struct hlist_node *tmp;
1036 struct hlist_head *bucket;
1041 /* Note that we do not need to protect our mac_hash traversal with rcu,
1042 * since all modification code is protected by mdev->state_lock
1045 /* find what to remove */
1046 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1047 bucket = &priv->mac_hash[i];
1048 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1050 netdev_for_each_uc_addr(ha, dev) {
1051 if (ether_addr_equal_64bits(entry->mac,
1058 /* MAC address of the port is not in uc list */
1059 if (ether_addr_equal_64bits(entry->mac,
1064 mac = mlx4_mac_to_u64(entry->mac);
1065 mlx4_en_uc_steer_release(priv, entry->mac,
1068 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1070 hlist_del_rcu(&entry->hlist);
1071 kfree_rcu(entry, rcu);
1072 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1073 entry->mac, priv->port);
1079 /* if we didn't remove anything, there is no use in trying to add
1080 * again once we are in a forced promisc mode state
1082 if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1085 prev_flags = priv->flags;
1086 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1088 /* find what to add */
1089 netdev_for_each_uc_addr(ha, dev) {
1091 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1092 hlist_for_each_entry(entry, bucket, hlist) {
1093 if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1100 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1102 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1103 ha->addr, priv->port);
1104 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1107 mac = mlx4_mac_to_u64(ha->addr);
1108 memcpy(entry->mac, ha->addr, ETH_ALEN);
1109 err = mlx4_register_mac(mdev->dev, priv->port, mac);
1111 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1112 ha->addr, priv->port, err);
1114 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1117 err = mlx4_en_uc_steer_add(priv, ha->addr,
1121 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1122 ha->addr, priv->port, err);
1123 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1125 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1128 unsigned int mac_hash;
1129 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1130 ha->addr, priv->port);
1131 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1132 bucket = &priv->mac_hash[mac_hash];
1133 hlist_add_head_rcu(&entry->hlist, bucket);
1138 if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1139 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1141 } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1142 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1147 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1149 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1151 struct mlx4_en_dev *mdev = priv->mdev;
1152 struct net_device *dev = priv->dev;
1154 mutex_lock(&mdev->state_lock);
1155 if (!mdev->device_up) {
1156 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1159 if (!priv->port_up) {
1160 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1164 if (!netif_carrier_ok(dev)) {
1165 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1166 if (priv->port_state.link_state) {
1167 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1168 netif_carrier_on(dev);
1169 en_dbg(LINK, priv, "Link Up\n");
1174 if (dev->priv_flags & IFF_UNICAST_FLT)
1175 mlx4_en_do_uc_filter(priv, dev, mdev);
1177 /* Promsicuous mode: disable all filters */
1178 if ((dev->flags & IFF_PROMISC) ||
1179 (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1180 mlx4_en_set_promisc_mode(priv, mdev);
1184 /* Not in promiscuous mode */
1185 if (priv->flags & MLX4_EN_FLAG_PROMISC)
1186 mlx4_en_clear_promisc_mode(priv, mdev);
1188 mlx4_en_do_multicast(priv, dev, mdev);
1190 mutex_unlock(&mdev->state_lock);
1193 #ifdef CONFIG_NET_POLL_CONTROLLER
1194 static void mlx4_en_netpoll(struct net_device *dev)
1196 struct mlx4_en_priv *priv = netdev_priv(dev);
1197 struct mlx4_en_cq *cq;
1200 for (i = 0; i < priv->rx_ring_num; i++) {
1201 cq = priv->rx_cq[i];
1202 napi_schedule(&cq->napi);
1207 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1211 int *qpn = &priv->base_qpn;
1212 struct mlx4_mac_entry *entry;
1214 err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, ®_id);
1218 err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1219 &priv->tunnel_reg_id);
1223 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1229 memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1230 memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1231 entry->reg_id = reg_id;
1232 hlist_add_head_rcu(&entry->hlist,
1233 &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1238 if (priv->tunnel_reg_id)
1239 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1242 mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1246 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1250 int qpn = priv->base_qpn;
1251 struct hlist_head *bucket;
1252 struct hlist_node *tmp;
1253 struct mlx4_mac_entry *entry;
1255 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1256 bucket = &priv->mac_hash[i];
1257 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1258 mac = mlx4_mac_to_u64(entry->mac);
1259 en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1261 mlx4_en_uc_steer_release(priv, entry->mac,
1262 qpn, entry->reg_id);
1264 mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1265 hlist_del_rcu(&entry->hlist);
1266 kfree_rcu(entry, rcu);
1270 if (priv->tunnel_reg_id) {
1271 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1272 priv->tunnel_reg_id = 0;
1276 static void mlx4_en_tx_timeout(struct net_device *dev)
1278 struct mlx4_en_priv *priv = netdev_priv(dev);
1279 struct mlx4_en_dev *mdev = priv->mdev;
1282 if (netif_msg_timer(priv))
1283 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1285 for (i = 0; i < priv->tx_ring_num; i++) {
1286 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1288 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1289 i, priv->tx_ring[i]->qpn, priv->tx_ring[i]->cqn,
1290 priv->tx_ring[i]->cons, priv->tx_ring[i]->prod);
1293 priv->port_stats.tx_timeout++;
1294 en_dbg(DRV, priv, "Scheduling watchdog\n");
1295 queue_work(mdev->workqueue, &priv->watchdog_task);
1299 static struct rtnl_link_stats64 *
1300 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1302 struct mlx4_en_priv *priv = netdev_priv(dev);
1304 spin_lock_bh(&priv->stats_lock);
1305 netdev_stats_to_stats64(stats, &dev->stats);
1306 spin_unlock_bh(&priv->stats_lock);
1311 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1313 struct mlx4_en_cq *cq;
1316 /* If we haven't received a specific coalescing setting
1317 * (module param), we set the moderation parameters as follows:
1318 * - moder_cnt is set to the number of mtu sized packets to
1319 * satisfy our coalescing target.
1320 * - moder_time is set to a fixed value.
1322 priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1323 priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1324 priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1325 priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1326 en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1327 priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1329 /* Setup cq moderation params */
1330 for (i = 0; i < priv->rx_ring_num; i++) {
1331 cq = priv->rx_cq[i];
1332 cq->moder_cnt = priv->rx_frames;
1333 cq->moder_time = priv->rx_usecs;
1334 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1335 priv->last_moder_packets[i] = 0;
1336 priv->last_moder_bytes[i] = 0;
1339 for (i = 0; i < priv->tx_ring_num; i++) {
1340 cq = priv->tx_cq[i];
1341 cq->moder_cnt = priv->tx_frames;
1342 cq->moder_time = priv->tx_usecs;
1345 /* Reset auto-moderation params */
1346 priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1347 priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1348 priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1349 priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1350 priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1351 priv->adaptive_rx_coal = 1;
1352 priv->last_moder_jiffies = 0;
1353 priv->last_moder_tx_packets = 0;
1356 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1358 unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1359 struct mlx4_en_cq *cq;
1360 unsigned long packets;
1362 unsigned long avg_pkt_size;
1363 unsigned long rx_packets;
1364 unsigned long rx_bytes;
1365 unsigned long rx_pkt_diff;
1369 if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1372 for (ring = 0; ring < priv->rx_ring_num; ring++) {
1373 spin_lock_bh(&priv->stats_lock);
1374 rx_packets = priv->rx_ring[ring]->packets;
1375 rx_bytes = priv->rx_ring[ring]->bytes;
1376 spin_unlock_bh(&priv->stats_lock);
1378 rx_pkt_diff = ((unsigned long) (rx_packets -
1379 priv->last_moder_packets[ring]));
1380 packets = rx_pkt_diff;
1381 rate = packets * HZ / period;
1382 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1383 priv->last_moder_bytes[ring])) / packets : 0;
1385 /* Apply auto-moderation only when packet rate
1386 * exceeds a rate that it matters */
1387 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1388 avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1389 if (rate < priv->pkt_rate_low)
1390 moder_time = priv->rx_usecs_low;
1391 else if (rate > priv->pkt_rate_high)
1392 moder_time = priv->rx_usecs_high;
1394 moder_time = (rate - priv->pkt_rate_low) *
1395 (priv->rx_usecs_high - priv->rx_usecs_low) /
1396 (priv->pkt_rate_high - priv->pkt_rate_low) +
1399 moder_time = priv->rx_usecs_low;
1402 if (moder_time != priv->last_moder_time[ring]) {
1403 priv->last_moder_time[ring] = moder_time;
1404 cq = priv->rx_cq[ring];
1405 cq->moder_time = moder_time;
1406 cq->moder_cnt = priv->rx_frames;
1407 err = mlx4_en_set_cq_moder(priv, cq);
1409 en_err(priv, "Failed modifying moderation for cq:%d\n",
1412 priv->last_moder_packets[ring] = rx_packets;
1413 priv->last_moder_bytes[ring] = rx_bytes;
1416 priv->last_moder_jiffies = jiffies;
1419 static void mlx4_en_do_get_stats(struct work_struct *work)
1421 struct delayed_work *delay = to_delayed_work(work);
1422 struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1424 struct mlx4_en_dev *mdev = priv->mdev;
1427 mutex_lock(&mdev->state_lock);
1428 if (mdev->device_up) {
1429 if (priv->port_up) {
1430 err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1432 en_dbg(HW, priv, "Could not update stats\n");
1434 mlx4_en_auto_moderation(priv);
1437 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1439 if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1440 mlx4_en_do_set_mac(priv, priv->current_mac);
1441 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1443 mutex_unlock(&mdev->state_lock);
1446 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1449 static void mlx4_en_service_task(struct work_struct *work)
1451 struct delayed_work *delay = to_delayed_work(work);
1452 struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1454 struct mlx4_en_dev *mdev = priv->mdev;
1456 mutex_lock(&mdev->state_lock);
1457 if (mdev->device_up) {
1458 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1459 mlx4_en_ptp_overflow_check(mdev);
1461 mlx4_en_recover_from_oom(priv);
1462 queue_delayed_work(mdev->workqueue, &priv->service_task,
1463 SERVICE_TASK_DELAY);
1465 mutex_unlock(&mdev->state_lock);
1468 static void mlx4_en_linkstate(struct work_struct *work)
1470 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1472 struct mlx4_en_dev *mdev = priv->mdev;
1473 int linkstate = priv->link_state;
1475 mutex_lock(&mdev->state_lock);
1476 /* If observable port state changed set carrier state and
1477 * report to system log */
1478 if (priv->last_link_state != linkstate) {
1479 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1480 en_info(priv, "Link Down\n");
1481 netif_carrier_off(priv->dev);
1483 en_info(priv, "Link Up\n");
1484 netif_carrier_on(priv->dev);
1487 priv->last_link_state = linkstate;
1488 mutex_unlock(&mdev->state_lock);
1491 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1493 struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1494 int numa_node = priv->mdev->dev->numa_node;
1496 if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1499 cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1500 ring->affinity_mask);
1504 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1506 free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1509 int mlx4_en_start_port(struct net_device *dev)
1511 struct mlx4_en_priv *priv = netdev_priv(dev);
1512 struct mlx4_en_dev *mdev = priv->mdev;
1513 struct mlx4_en_cq *cq;
1514 struct mlx4_en_tx_ring *tx_ring;
1520 u8 mc_list[16] = {0};
1522 if (priv->port_up) {
1523 en_dbg(DRV, priv, "start port called while port already up\n");
1527 INIT_LIST_HEAD(&priv->mc_list);
1528 INIT_LIST_HEAD(&priv->curr_list);
1529 INIT_LIST_HEAD(&priv->ethtool_list);
1530 memset(&priv->ethtool_rules[0], 0,
1531 sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1533 /* Calculate Rx buf size */
1534 dev->mtu = min(dev->mtu, priv->max_mtu);
1535 mlx4_en_calc_rx_buf(dev);
1536 en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1538 /* Configure rx cq's and rings */
1539 err = mlx4_en_activate_rx_rings(priv);
1541 en_err(priv, "Failed to activate RX rings\n");
1544 for (i = 0; i < priv->rx_ring_num; i++) {
1545 cq = priv->rx_cq[i];
1547 err = mlx4_en_init_affinity_hint(priv, i);
1549 en_err(priv, "Failed preparing IRQ affinity hint\n");
1553 err = mlx4_en_activate_cq(priv, cq, i);
1555 en_err(priv, "Failed activating Rx CQ\n");
1556 mlx4_en_free_affinity_hint(priv, i);
1560 for (j = 0; j < cq->size; j++) {
1561 struct mlx4_cqe *cqe = NULL;
1563 cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1565 cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1568 err = mlx4_en_set_cq_moder(priv, cq);
1570 en_err(priv, "Failed setting cq moderation parameters\n");
1571 mlx4_en_deactivate_cq(priv, cq);
1572 mlx4_en_free_affinity_hint(priv, i);
1575 mlx4_en_arm_cq(priv, cq);
1576 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1581 en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1582 err = mlx4_en_get_qp(priv);
1584 en_err(priv, "Failed getting eth qp\n");
1587 mdev->mac_removed[priv->port] = 0;
1589 priv->counter_index =
1590 mlx4_get_default_counter_index(mdev->dev, priv->port);
1592 err = mlx4_en_config_rss_steer(priv);
1594 en_err(priv, "Failed configuring rss steering\n");
1598 err = mlx4_en_create_drop_qp(priv);
1602 /* Configure tx cq's and rings */
1603 for (i = 0; i < priv->tx_ring_num; i++) {
1605 cq = priv->tx_cq[i];
1606 err = mlx4_en_activate_cq(priv, cq, i);
1608 en_err(priv, "Failed allocating Tx CQ\n");
1611 err = mlx4_en_set_cq_moder(priv, cq);
1613 en_err(priv, "Failed setting cq moderation parameters\n");
1614 mlx4_en_deactivate_cq(priv, cq);
1617 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1618 cq->buf->wqe_index = cpu_to_be16(0xffff);
1620 /* Configure ring */
1621 tx_ring = priv->tx_ring[i];
1622 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1623 i / priv->num_tx_rings_p_up);
1625 en_err(priv, "Failed allocating Tx ring\n");
1626 mlx4_en_deactivate_cq(priv, cq);
1629 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1631 /* Arm CQ for TX completions */
1632 mlx4_en_arm_cq(priv, cq);
1634 /* Set initial ownership of all Tx TXBBs to SW (1) */
1635 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1636 *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1640 /* Configure port */
1641 err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1642 priv->rx_skb_size + ETH_FCS_LEN,
1643 priv->prof->tx_pause,
1645 priv->prof->rx_pause,
1646 priv->prof->rx_ppp);
1648 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1652 /* Set default qp number */
1653 err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1655 en_err(priv, "Failed setting default qp numbers\n");
1659 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1660 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1662 en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1669 en_dbg(HW, priv, "Initializing port\n");
1670 err = mlx4_INIT_PORT(mdev->dev, priv->port);
1672 en_err(priv, "Failed Initializing port\n");
1676 /* Set Unicast and VXLAN steering rules */
1677 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1678 mlx4_en_set_rss_steer_rules(priv))
1679 mlx4_warn(mdev, "Failed setting steering rules\n");
1681 /* Attach rx QP to bradcast address */
1682 eth_broadcast_addr(&mc_list[10]);
1683 mc_list[5] = priv->port; /* needed for B0 steering support */
1684 if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1685 priv->port, 0, MLX4_PROT_ETH,
1686 &priv->broadcast_id))
1687 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1689 /* Must redo promiscuous mode setup. */
1690 priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1692 /* Schedule multicast task to populate multicast list */
1693 queue_work(mdev->workqueue, &priv->rx_mode_task);
1695 #ifdef CONFIG_MLX4_EN_VXLAN
1696 if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1697 vxlan_get_rx_port(dev);
1699 priv->port_up = true;
1700 netif_tx_start_all_queues(dev);
1701 netif_device_attach(dev);
1706 while (tx_index--) {
1707 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1708 mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1710 mlx4_en_destroy_drop_qp(priv);
1712 mlx4_en_release_rss_steer(priv);
1714 mlx4_en_put_qp(priv);
1716 while (rx_index--) {
1717 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1718 mlx4_en_free_affinity_hint(priv, rx_index);
1720 for (i = 0; i < priv->rx_ring_num; i++)
1721 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1723 return err; /* need to close devices */
1727 void mlx4_en_stop_port(struct net_device *dev, int detach)
1729 struct mlx4_en_priv *priv = netdev_priv(dev);
1730 struct mlx4_en_dev *mdev = priv->mdev;
1731 struct mlx4_en_mc_list *mclist, *tmp;
1732 struct ethtool_flow_id *flow, *tmp_flow;
1734 u8 mc_list[16] = {0};
1736 if (!priv->port_up) {
1737 en_dbg(DRV, priv, "stop port called while port already down\n");
1742 mlx4_CLOSE_PORT(mdev->dev, priv->port);
1744 /* Synchronize with tx routine */
1745 netif_tx_lock_bh(dev);
1747 netif_device_detach(dev);
1748 netif_tx_stop_all_queues(dev);
1749 netif_tx_unlock_bh(dev);
1751 netif_tx_disable(dev);
1753 /* Set port as not active */
1754 priv->port_up = false;
1755 priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1757 /* Promsicuous mode */
1758 if (mdev->dev->caps.steering_mode ==
1759 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1760 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1761 MLX4_EN_FLAG_MC_PROMISC);
1762 mlx4_flow_steer_promisc_remove(mdev->dev,
1764 MLX4_FS_ALL_DEFAULT);
1765 mlx4_flow_steer_promisc_remove(mdev->dev,
1767 MLX4_FS_MC_DEFAULT);
1768 } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1769 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1771 /* Disable promiscouos mode */
1772 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1775 /* Disable Multicast promisc */
1776 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1777 mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1779 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1783 /* Detach All multicasts */
1784 eth_broadcast_addr(&mc_list[10]);
1785 mc_list[5] = priv->port; /* needed for B0 steering support */
1786 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1787 MLX4_PROT_ETH, priv->broadcast_id);
1788 list_for_each_entry(mclist, &priv->curr_list, list) {
1789 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1790 mc_list[5] = priv->port;
1791 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1792 mc_list, MLX4_PROT_ETH, mclist->reg_id);
1793 if (mclist->tunnel_reg_id)
1794 mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1796 mlx4_en_clear_list(dev);
1797 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1798 list_del(&mclist->list);
1802 /* Flush multicast filter */
1803 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1805 /* Remove flow steering rules for the port*/
1806 if (mdev->dev->caps.steering_mode ==
1807 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1809 list_for_each_entry_safe(flow, tmp_flow,
1810 &priv->ethtool_list, list) {
1811 mlx4_flow_detach(mdev->dev, flow->id);
1812 list_del(&flow->list);
1816 mlx4_en_destroy_drop_qp(priv);
1819 for (i = 0; i < priv->tx_ring_num; i++) {
1820 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1821 mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1825 for (i = 0; i < priv->tx_ring_num; i++)
1826 mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1828 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1829 mlx4_en_delete_rss_steer_rules(priv);
1832 mlx4_en_release_rss_steer(priv);
1834 /* Unregister Mac address for the port */
1835 mlx4_en_put_qp(priv);
1836 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1837 mdev->mac_removed[priv->port] = 1;
1840 for (i = 0; i < priv->rx_ring_num; i++) {
1841 struct mlx4_en_cq *cq = priv->rx_cq[i];
1843 napi_synchronize(&cq->napi);
1844 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1845 mlx4_en_deactivate_cq(priv, cq);
1847 mlx4_en_free_affinity_hint(priv, i);
1851 static void mlx4_en_restart(struct work_struct *work)
1853 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1855 struct mlx4_en_dev *mdev = priv->mdev;
1856 struct net_device *dev = priv->dev;
1858 en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1861 mutex_lock(&mdev->state_lock);
1862 if (priv->port_up) {
1863 mlx4_en_stop_port(dev, 1);
1864 if (mlx4_en_start_port(dev))
1865 en_err(priv, "Failed restarting port %d\n", priv->port);
1867 mutex_unlock(&mdev->state_lock);
1871 static void mlx4_en_clear_stats(struct net_device *dev)
1873 struct mlx4_en_priv *priv = netdev_priv(dev);
1874 struct mlx4_en_dev *mdev = priv->mdev;
1877 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1878 en_dbg(HW, priv, "Failed dumping statistics\n");
1880 memset(&priv->pstats, 0, sizeof(priv->pstats));
1881 memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1882 memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1883 memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
1884 memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
1885 memset(&priv->rx_priority_flowstats, 0,
1886 sizeof(priv->rx_priority_flowstats));
1887 memset(&priv->tx_priority_flowstats, 0,
1888 sizeof(priv->tx_priority_flowstats));
1889 memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
1891 for (i = 0; i < priv->tx_ring_num; i++) {
1892 priv->tx_ring[i]->bytes = 0;
1893 priv->tx_ring[i]->packets = 0;
1894 priv->tx_ring[i]->tx_csum = 0;
1895 priv->tx_ring[i]->tx_dropped = 0;
1896 priv->tx_ring[i]->queue_stopped = 0;
1897 priv->tx_ring[i]->wake_queue = 0;
1898 priv->tx_ring[i]->tso_packets = 0;
1899 priv->tx_ring[i]->xmit_more = 0;
1901 for (i = 0; i < priv->rx_ring_num; i++) {
1902 priv->rx_ring[i]->bytes = 0;
1903 priv->rx_ring[i]->packets = 0;
1904 priv->rx_ring[i]->csum_ok = 0;
1905 priv->rx_ring[i]->csum_none = 0;
1906 priv->rx_ring[i]->csum_complete = 0;
1910 static int mlx4_en_open(struct net_device *dev)
1912 struct mlx4_en_priv *priv = netdev_priv(dev);
1913 struct mlx4_en_dev *mdev = priv->mdev;
1916 mutex_lock(&mdev->state_lock);
1918 if (!mdev->device_up) {
1919 en_err(priv, "Cannot open - device down/disabled\n");
1924 /* Reset HW statistics and SW counters */
1925 mlx4_en_clear_stats(dev);
1927 err = mlx4_en_start_port(dev);
1929 en_err(priv, "Failed starting port:%d\n", priv->port);
1932 mutex_unlock(&mdev->state_lock);
1937 static int mlx4_en_close(struct net_device *dev)
1939 struct mlx4_en_priv *priv = netdev_priv(dev);
1940 struct mlx4_en_dev *mdev = priv->mdev;
1942 en_dbg(IFDOWN, priv, "Close port called\n");
1944 mutex_lock(&mdev->state_lock);
1946 mlx4_en_stop_port(dev, 0);
1947 netif_carrier_off(dev);
1949 mutex_unlock(&mdev->state_lock);
1953 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1957 #ifdef CONFIG_RFS_ACCEL
1958 priv->dev->rx_cpu_rmap = NULL;
1961 for (i = 0; i < priv->tx_ring_num; i++) {
1962 if (priv->tx_ring && priv->tx_ring[i])
1963 mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1964 if (priv->tx_cq && priv->tx_cq[i])
1965 mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1968 for (i = 0; i < priv->rx_ring_num; i++) {
1969 if (priv->rx_ring[i])
1970 mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1971 priv->prof->rx_ring_size, priv->stride);
1973 mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1978 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1980 struct mlx4_en_port_profile *prof = priv->prof;
1984 /* Create tx Rings */
1985 for (i = 0; i < priv->tx_ring_num; i++) {
1986 node = cpu_to_node(i % num_online_cpus());
1987 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1988 prof->tx_ring_size, i, TX, node))
1991 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
1992 prof->tx_ring_size, TXBB_SIZE,
1997 /* Create rx Rings */
1998 for (i = 0; i < priv->rx_ring_num; i++) {
1999 node = cpu_to_node(i % num_online_cpus());
2000 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2001 prof->rx_ring_size, i, RX, node))
2004 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2005 prof->rx_ring_size, priv->stride,
2010 #ifdef CONFIG_RFS_ACCEL
2011 priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2017 en_err(priv, "Failed to allocate NIC resources\n");
2018 for (i = 0; i < priv->rx_ring_num; i++) {
2019 if (priv->rx_ring[i])
2020 mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2024 mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2026 for (i = 0; i < priv->tx_ring_num; i++) {
2027 if (priv->tx_ring[i])
2028 mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
2030 mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
2036 void mlx4_en_destroy_netdev(struct net_device *dev)
2038 struct mlx4_en_priv *priv = netdev_priv(dev);
2039 struct mlx4_en_dev *mdev = priv->mdev;
2041 en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2043 /* Unregister device - this will close the port if it was up */
2044 if (priv->registered) {
2045 devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2047 unregister_netdev(dev);
2050 if (priv->allocated)
2051 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2053 cancel_delayed_work(&priv->stats_task);
2054 cancel_delayed_work(&priv->service_task);
2055 /* flush any pending task for this netdev */
2056 flush_workqueue(mdev->workqueue);
2058 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2059 mlx4_en_remove_timestamp(mdev);
2061 /* Detach the netdev so tasks would not attempt to access it */
2062 mutex_lock(&mdev->state_lock);
2063 mdev->pndev[priv->port] = NULL;
2064 mdev->upper[priv->port] = NULL;
2065 mutex_unlock(&mdev->state_lock);
2067 mlx4_en_free_resources(priv);
2069 kfree(priv->tx_ring);
2075 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2077 struct mlx4_en_priv *priv = netdev_priv(dev);
2078 struct mlx4_en_dev *mdev = priv->mdev;
2081 en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2084 if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
2085 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
2090 if (netif_running(dev)) {
2091 mutex_lock(&mdev->state_lock);
2092 if (!mdev->device_up) {
2093 /* NIC is probably restarting - let watchdog task reset
2095 en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2097 mlx4_en_stop_port(dev, 1);
2098 err = mlx4_en_start_port(dev);
2100 en_err(priv, "Failed restarting port:%d\n",
2102 queue_work(mdev->workqueue, &priv->watchdog_task);
2105 mutex_unlock(&mdev->state_lock);
2110 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2112 struct mlx4_en_priv *priv = netdev_priv(dev);
2113 struct mlx4_en_dev *mdev = priv->mdev;
2114 struct hwtstamp_config config;
2116 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2119 /* reserved for future extensions */
2123 /* device doesn't support time stamping */
2124 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2127 /* TX HW timestamp */
2128 switch (config.tx_type) {
2129 case HWTSTAMP_TX_OFF:
2130 case HWTSTAMP_TX_ON:
2136 /* RX HW timestamp */
2137 switch (config.rx_filter) {
2138 case HWTSTAMP_FILTER_NONE:
2140 case HWTSTAMP_FILTER_ALL:
2141 case HWTSTAMP_FILTER_SOME:
2142 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2143 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2144 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2145 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2146 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2147 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2148 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2149 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2150 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2151 case HWTSTAMP_FILTER_PTP_V2_EVENT:
2152 case HWTSTAMP_FILTER_PTP_V2_SYNC:
2153 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2154 config.rx_filter = HWTSTAMP_FILTER_ALL;
2160 if (mlx4_en_reset_config(dev, config, dev->features)) {
2161 config.tx_type = HWTSTAMP_TX_OFF;
2162 config.rx_filter = HWTSTAMP_FILTER_NONE;
2165 return copy_to_user(ifr->ifr_data, &config,
2166 sizeof(config)) ? -EFAULT : 0;
2169 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2171 struct mlx4_en_priv *priv = netdev_priv(dev);
2173 return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2174 sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2177 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2181 return mlx4_en_hwtstamp_set(dev, ifr);
2183 return mlx4_en_hwtstamp_get(dev, ifr);
2189 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2190 netdev_features_t features)
2192 struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2193 struct mlx4_en_dev *mdev = en_priv->mdev;
2195 /* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2196 * enable/disable make sure S-TAG flag is always in same state as
2199 if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2200 !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2201 features |= NETIF_F_HW_VLAN_STAG_RX;
2203 features &= ~NETIF_F_HW_VLAN_STAG_RX;
2208 static int mlx4_en_set_features(struct net_device *netdev,
2209 netdev_features_t features)
2211 struct mlx4_en_priv *priv = netdev_priv(netdev);
2215 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2216 en_info(priv, "Turn %s RX-FCS\n",
2217 (features & NETIF_F_RXFCS) ? "ON" : "OFF");
2221 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2222 u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2224 en_info(priv, "Turn %s RX-ALL\n",
2225 ignore_fcs_value ? "ON" : "OFF");
2226 ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2227 priv->port, ignore_fcs_value);
2232 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2233 en_info(priv, "Turn %s RX vlan strip offload\n",
2234 (features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2238 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2239 en_info(priv, "Turn %s TX vlan strip offload\n",
2240 (features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2242 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2243 en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2244 (features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2246 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2247 en_info(priv, "Turn %s loopback\n",
2248 (features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2249 mlx4_en_update_loopback_state(netdev, features);
2253 ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2262 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2264 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2265 struct mlx4_en_dev *mdev = en_priv->mdev;
2266 u64 mac_u64 = mlx4_mac_to_u64(mac);
2268 if (is_multicast_ether_addr(mac))
2271 return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2274 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos)
2276 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2277 struct mlx4_en_dev *mdev = en_priv->mdev;
2279 return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos);
2282 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2285 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2286 struct mlx4_en_dev *mdev = en_priv->mdev;
2288 return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2292 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2294 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2295 struct mlx4_en_dev *mdev = en_priv->mdev;
2297 return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2300 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2302 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2303 struct mlx4_en_dev *mdev = en_priv->mdev;
2305 return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2308 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2310 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2311 struct mlx4_en_dev *mdev = en_priv->mdev;
2313 return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2316 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2317 struct ifla_vf_stats *vf_stats)
2319 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2320 struct mlx4_en_dev *mdev = en_priv->mdev;
2322 return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2325 #define PORT_ID_BYTE_LEN 8
2326 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2327 struct netdev_phys_item_id *ppid)
2329 struct mlx4_en_priv *priv = netdev_priv(dev);
2330 struct mlx4_dev *mdev = priv->mdev->dev;
2332 u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2337 ppid->id_len = sizeof(phys_port_id);
2338 for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2339 ppid->id[i] = phys_port_id & 0xff;
2345 #ifdef CONFIG_MLX4_EN_VXLAN
2346 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2349 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2352 ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2356 ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2357 VXLAN_STEER_BY_OUTER_MAC, 1);
2360 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2365 priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2367 NETIF_F_TSO | NETIF_F_TSO6 |
2368 NETIF_F_GSO_UDP_TUNNEL |
2369 NETIF_F_GSO_UDP_TUNNEL_CSUM |
2370 NETIF_F_GSO_PARTIAL;
2373 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2376 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2378 /* unset offloads */
2379 priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2381 NETIF_F_TSO | NETIF_F_TSO6 |
2382 NETIF_F_GSO_UDP_TUNNEL |
2383 NETIF_F_GSO_UDP_TUNNEL_CSUM |
2384 NETIF_F_GSO_PARTIAL);
2386 ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2387 VXLAN_STEER_BY_OUTER_MAC, 0);
2389 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2391 priv->vxlan_port = 0;
2394 static void mlx4_en_add_vxlan_port(struct net_device *dev,
2395 sa_family_t sa_family, __be16 port)
2397 struct mlx4_en_priv *priv = netdev_priv(dev);
2398 __be16 current_port;
2400 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2403 if (sa_family == AF_INET6)
2406 current_port = priv->vxlan_port;
2407 if (current_port && current_port != port) {
2408 en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2409 ntohs(current_port), ntohs(port));
2413 priv->vxlan_port = port;
2414 queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2417 static void mlx4_en_del_vxlan_port(struct net_device *dev,
2418 sa_family_t sa_family, __be16 port)
2420 struct mlx4_en_priv *priv = netdev_priv(dev);
2421 __be16 current_port;
2423 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2426 if (sa_family == AF_INET6)
2429 current_port = priv->vxlan_port;
2430 if (current_port != port) {
2431 en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2435 queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2438 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2439 struct net_device *dev,
2440 netdev_features_t features)
2442 features = vlan_features_check(skb, features);
2443 features = vxlan_features_check(skb, features);
2445 /* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2446 * support inner IPv6 checksums and segmentation so we need to
2447 * strip that feature if this is an IPv6 encapsulated frame.
2449 if (skb->encapsulation &&
2450 (skb->ip_summed == CHECKSUM_PARTIAL) &&
2451 (ip_hdr(skb)->version != 4))
2452 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2458 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2460 struct mlx4_en_priv *priv = netdev_priv(dev);
2461 struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[queue_index];
2462 struct mlx4_update_qp_params params;
2465 if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2468 /* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2469 if (maxrate >> 12) {
2470 params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2471 params.rate_val = maxrate / 1000;
2472 } else if (maxrate) {
2473 params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2474 params.rate_val = maxrate;
2475 } else { /* zero serves to revoke the QP rate-limitation */
2476 params.rate_unit = 0;
2477 params.rate_val = 0;
2480 err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2485 static const struct net_device_ops mlx4_netdev_ops = {
2486 .ndo_open = mlx4_en_open,
2487 .ndo_stop = mlx4_en_close,
2488 .ndo_start_xmit = mlx4_en_xmit,
2489 .ndo_select_queue = mlx4_en_select_queue,
2490 .ndo_get_stats64 = mlx4_en_get_stats64,
2491 .ndo_set_rx_mode = mlx4_en_set_rx_mode,
2492 .ndo_set_mac_address = mlx4_en_set_mac,
2493 .ndo_validate_addr = eth_validate_addr,
2494 .ndo_change_mtu = mlx4_en_change_mtu,
2495 .ndo_do_ioctl = mlx4_en_ioctl,
2496 .ndo_tx_timeout = mlx4_en_tx_timeout,
2497 .ndo_vlan_rx_add_vid = mlx4_en_vlan_rx_add_vid,
2498 .ndo_vlan_rx_kill_vid = mlx4_en_vlan_rx_kill_vid,
2499 #ifdef CONFIG_NET_POLL_CONTROLLER
2500 .ndo_poll_controller = mlx4_en_netpoll,
2502 .ndo_set_features = mlx4_en_set_features,
2503 .ndo_fix_features = mlx4_en_fix_features,
2504 .ndo_setup_tc = __mlx4_en_setup_tc,
2505 #ifdef CONFIG_RFS_ACCEL
2506 .ndo_rx_flow_steer = mlx4_en_filter_rfs,
2508 .ndo_get_phys_port_id = mlx4_en_get_phys_port_id,
2509 #ifdef CONFIG_MLX4_EN_VXLAN
2510 .ndo_add_vxlan_port = mlx4_en_add_vxlan_port,
2511 .ndo_del_vxlan_port = mlx4_en_del_vxlan_port,
2512 .ndo_features_check = mlx4_en_features_check,
2514 .ndo_set_tx_maxrate = mlx4_en_set_tx_maxrate,
2517 static const struct net_device_ops mlx4_netdev_ops_master = {
2518 .ndo_open = mlx4_en_open,
2519 .ndo_stop = mlx4_en_close,
2520 .ndo_start_xmit = mlx4_en_xmit,
2521 .ndo_select_queue = mlx4_en_select_queue,
2522 .ndo_get_stats64 = mlx4_en_get_stats64,
2523 .ndo_set_rx_mode = mlx4_en_set_rx_mode,
2524 .ndo_set_mac_address = mlx4_en_set_mac,
2525 .ndo_validate_addr = eth_validate_addr,
2526 .ndo_change_mtu = mlx4_en_change_mtu,
2527 .ndo_tx_timeout = mlx4_en_tx_timeout,
2528 .ndo_vlan_rx_add_vid = mlx4_en_vlan_rx_add_vid,
2529 .ndo_vlan_rx_kill_vid = mlx4_en_vlan_rx_kill_vid,
2530 .ndo_set_vf_mac = mlx4_en_set_vf_mac,
2531 .ndo_set_vf_vlan = mlx4_en_set_vf_vlan,
2532 .ndo_set_vf_rate = mlx4_en_set_vf_rate,
2533 .ndo_set_vf_spoofchk = mlx4_en_set_vf_spoofchk,
2534 .ndo_set_vf_link_state = mlx4_en_set_vf_link_state,
2535 .ndo_get_vf_stats = mlx4_en_get_vf_stats,
2536 .ndo_get_vf_config = mlx4_en_get_vf_config,
2537 #ifdef CONFIG_NET_POLL_CONTROLLER
2538 .ndo_poll_controller = mlx4_en_netpoll,
2540 .ndo_set_features = mlx4_en_set_features,
2541 .ndo_fix_features = mlx4_en_fix_features,
2542 .ndo_setup_tc = __mlx4_en_setup_tc,
2543 #ifdef CONFIG_RFS_ACCEL
2544 .ndo_rx_flow_steer = mlx4_en_filter_rfs,
2546 .ndo_get_phys_port_id = mlx4_en_get_phys_port_id,
2547 #ifdef CONFIG_MLX4_EN_VXLAN
2548 .ndo_add_vxlan_port = mlx4_en_add_vxlan_port,
2549 .ndo_del_vxlan_port = mlx4_en_del_vxlan_port,
2550 .ndo_features_check = mlx4_en_features_check,
2552 .ndo_set_tx_maxrate = mlx4_en_set_tx_maxrate,
2555 struct mlx4_en_bond {
2556 struct work_struct work;
2557 struct mlx4_en_priv *priv;
2559 struct mlx4_port_map port_map;
2562 static void mlx4_en_bond_work(struct work_struct *work)
2564 struct mlx4_en_bond *bond = container_of(work,
2565 struct mlx4_en_bond,
2568 struct mlx4_dev *dev = bond->priv->mdev->dev;
2570 if (bond->is_bonded) {
2571 if (!mlx4_is_bonded(dev)) {
2572 err = mlx4_bond(dev);
2574 en_err(bond->priv, "Fail to bond device\n");
2577 err = mlx4_port_map_set(dev, &bond->port_map);
2579 en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2580 bond->port_map.port1,
2581 bond->port_map.port2,
2584 } else if (mlx4_is_bonded(dev)) {
2585 err = mlx4_unbond(dev);
2587 en_err(bond->priv, "Fail to unbond device\n");
2589 dev_put(bond->priv->dev);
2593 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2594 u8 v2p_p1, u8 v2p_p2)
2596 struct mlx4_en_bond *bond = NULL;
2598 bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2602 INIT_WORK(&bond->work, mlx4_en_bond_work);
2604 bond->is_bonded = is_bonded;
2605 bond->port_map.port1 = v2p_p1;
2606 bond->port_map.port2 = v2p_p2;
2607 dev_hold(priv->dev);
2608 queue_work(priv->mdev->workqueue, &bond->work);
2612 int mlx4_en_netdev_event(struct notifier_block *this,
2613 unsigned long event, void *ptr)
2615 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2617 struct mlx4_en_dev *mdev;
2618 struct mlx4_dev *dev;
2619 int i, num_eth_ports = 0;
2620 bool do_bond = true;
2621 struct mlx4_en_priv *priv;
2625 if (!net_eq(dev_net(ndev), &init_net))
2628 mdev = container_of(this, struct mlx4_en_dev, nb);
2631 /* Go into this mode only when two network devices set on two ports
2632 * of the same mlx4 device are slaves of the same bonding master
2634 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2636 if (!port && (mdev->pndev[i] == ndev))
2638 mdev->upper[i] = mdev->pndev[i] ?
2639 netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2640 /* condition not met: network device is a slave */
2641 if (!mdev->upper[i])
2643 if (num_eth_ports < 2)
2645 /* condition not met: same master */
2646 if (mdev->upper[i] != mdev->upper[i-1])
2649 /* condition not met: 2 salves */
2650 do_bond = (num_eth_ports == 2) ? do_bond : false;
2652 /* handle only events that come with enough info */
2653 if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
2656 priv = netdev_priv(ndev);
2658 struct netdev_notifier_bonding_info *notifier_info = ptr;
2659 struct netdev_bonding_info *bonding_info =
2660 ¬ifier_info->bonding_info;
2662 /* required mode 1, 2 or 4 */
2663 if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
2664 (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
2665 (bonding_info->master.bond_mode != BOND_MODE_8023AD))
2668 /* require exactly 2 slaves */
2669 if (bonding_info->master.num_slaves != 2)
2674 if (bonding_info->master.bond_mode ==
2675 BOND_MODE_ACTIVEBACKUP) {
2676 /* in active-backup mode virtual ports are
2677 * mapped to the physical port of the active
2679 if (bonding_info->slave.state ==
2680 BOND_STATE_BACKUP) {
2688 } else { /* BOND_STATE_ACTIVE */
2697 } else { /* Active-Active */
2698 /* in active-active mode a virtual port is
2699 * mapped to the native physical port if and only
2700 * if the physical port is up */
2701 __s8 link = bonding_info->slave.link;
2707 if ((link == BOND_LINK_UP) ||
2708 (link == BOND_LINK_FAIL)) {
2713 } else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
2723 mlx4_en_queue_bond_work(priv, do_bond,
2724 v2p_port1, v2p_port2);
2729 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
2730 struct mlx4_en_stats_bitmap *stats_bitmap,
2731 u8 rx_ppp, u8 rx_pause,
2732 u8 tx_ppp, u8 tx_pause)
2734 int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
2736 if (!mlx4_is_slave(dev) &&
2737 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
2738 mutex_lock(&stats_bitmap->mutex);
2739 bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
2742 bitmap_set(stats_bitmap->bitmap, last_i,
2743 NUM_FLOW_PRIORITY_STATS_RX);
2744 last_i += NUM_FLOW_PRIORITY_STATS_RX;
2746 if (rx_pause && !(rx_ppp))
2747 bitmap_set(stats_bitmap->bitmap, last_i,
2749 last_i += NUM_FLOW_STATS_RX;
2752 bitmap_set(stats_bitmap->bitmap, last_i,
2753 NUM_FLOW_PRIORITY_STATS_TX);
2754 last_i += NUM_FLOW_PRIORITY_STATS_TX;
2756 if (tx_pause && !(tx_ppp))
2757 bitmap_set(stats_bitmap->bitmap, last_i,
2759 last_i += NUM_FLOW_STATS_TX;
2761 mutex_unlock(&stats_bitmap->mutex);
2765 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
2766 struct mlx4_en_stats_bitmap *stats_bitmap,
2767 u8 rx_ppp, u8 rx_pause,
2768 u8 tx_ppp, u8 tx_pause)
2772 mutex_init(&stats_bitmap->mutex);
2773 bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
2775 if (mlx4_is_slave(dev)) {
2776 bitmap_set(stats_bitmap->bitmap, last_i +
2777 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
2778 bitmap_set(stats_bitmap->bitmap, last_i +
2779 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
2780 bitmap_set(stats_bitmap->bitmap, last_i +
2781 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
2782 bitmap_set(stats_bitmap->bitmap, last_i +
2783 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
2784 bitmap_set(stats_bitmap->bitmap, last_i +
2785 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
2786 bitmap_set(stats_bitmap->bitmap, last_i +
2787 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
2789 bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
2791 last_i += NUM_MAIN_STATS;
2793 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
2794 last_i += NUM_PORT_STATS;
2796 if (mlx4_is_master(dev))
2797 bitmap_set(stats_bitmap->bitmap, last_i,
2799 last_i += NUM_PF_STATS;
2801 mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
2804 last_i += NUM_FLOW_STATS;
2806 if (!mlx4_is_slave(dev))
2807 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
2810 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2811 struct mlx4_en_port_profile *prof)
2813 struct net_device *dev;
2814 struct mlx4_en_priv *priv;
2818 dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2819 MAX_TX_RINGS, MAX_RX_RINGS);
2823 netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2824 netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2826 SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
2827 dev->dev_port = port - 1;
2830 * Initialize driver private data
2833 priv = netdev_priv(dev);
2834 memset(priv, 0, sizeof(struct mlx4_en_priv));
2835 priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
2836 spin_lock_init(&priv->stats_lock);
2837 INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2838 INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2839 INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2840 INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2841 INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2842 #ifdef CONFIG_MLX4_EN_VXLAN
2843 INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
2844 INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
2846 #ifdef CONFIG_RFS_ACCEL
2847 INIT_LIST_HEAD(&priv->filters);
2848 spin_lock_init(&priv->filters_lock);
2853 priv->ddev = &mdev->pdev->dev;
2856 priv->port_up = false;
2857 priv->flags = prof->flags;
2858 priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
2859 priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2860 MLX4_WQE_CTRL_SOLICITED);
2861 priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2862 priv->tx_ring_num = prof->tx_ring_num;
2863 priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
2864 netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
2866 priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2868 if (!priv->tx_ring) {
2872 priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2878 priv->rx_ring_num = prof->rx_ring_num;
2879 priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2880 priv->cqe_size = mdev->dev->caps.cqe_size;
2881 priv->mac_index = -1;
2882 priv->msg_enable = MLX4_EN_MSG_LEVEL;
2883 #ifdef CONFIG_MLX4_EN_DCB
2884 if (!mlx4_is_slave(priv->mdev->dev)) {
2885 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
2886 dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2888 en_info(priv, "enabling only PFC DCB ops\n");
2889 dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2894 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2895 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2897 /* Query for default mac and max mtu */
2898 priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2900 if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
2901 MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
2902 priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
2904 /* Set default MAC */
2905 dev->addr_len = ETH_ALEN;
2906 mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2907 if (!is_valid_ether_addr(dev->dev_addr)) {
2908 en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2909 priv->port, dev->dev_addr);
2912 } else if (mlx4_is_slave(priv->mdev->dev) &&
2913 (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
2914 /* Random MAC was assigned in mlx4_slave_cap
2915 * in mlx4_core module
2917 dev->addr_assign_type |= NET_ADDR_RANDOM;
2918 en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
2921 memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
2923 priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2924 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2925 err = mlx4_en_alloc_resources(priv);
2929 /* Initialize time stamping config */
2930 priv->hwtstamp_config.flags = 0;
2931 priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
2932 priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2934 /* Allocate page for receive rings */
2935 err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2938 en_err(priv, "Failed to allocate page for rx qps\n");
2941 priv->allocated = 1;
2944 * Initialize netdev entry points
2946 if (mlx4_is_master(priv->mdev->dev))
2947 dev->netdev_ops = &mlx4_netdev_ops_master;
2949 dev->netdev_ops = &mlx4_netdev_ops;
2950 dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2951 netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2952 netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2954 dev->ethtool_ops = &mlx4_en_ethtool_ops;
2957 * Set driver features
2959 dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2960 if (mdev->LSO_support)
2961 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2963 dev->vlan_features = dev->hw_features;
2965 dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2966 dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2967 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2968 NETIF_F_HW_VLAN_CTAG_FILTER;
2969 dev->hw_features |= NETIF_F_LOOPBACK |
2970 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2972 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
2973 dev->features |= NETIF_F_HW_VLAN_STAG_RX |
2974 NETIF_F_HW_VLAN_STAG_FILTER;
2975 dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
2978 if (mlx4_is_slave(mdev->dev)) {
2981 err = get_phv_bit(mdev->dev, port, &phv);
2983 dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
2984 priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
2987 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
2988 !(mdev->dev->caps.flags2 &
2989 MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2990 dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
2993 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
2994 dev->hw_features |= NETIF_F_RXFCS;
2996 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
2997 dev->hw_features |= NETIF_F_RXALL;
2999 if (mdev->dev->caps.steering_mode ==
3000 MLX4_STEERING_MODE_DEVICE_MANAGED &&
3001 mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3002 dev->hw_features |= NETIF_F_NTUPLE;
3004 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3005 dev->priv_flags |= IFF_UNICAST_FLT;
3007 /* Setting a default hash function value */
3008 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3009 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3010 } else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3011 priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3014 "No RSS hash capabilities exposed, using Toeplitz\n");
3015 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3018 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3019 dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3020 NETIF_F_GSO_UDP_TUNNEL_CSUM |
3021 NETIF_F_GSO_PARTIAL;
3022 dev->features |= NETIF_F_GSO_UDP_TUNNEL |
3023 NETIF_F_GSO_UDP_TUNNEL_CSUM |
3024 NETIF_F_GSO_PARTIAL;
3025 dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3028 mdev->pndev[port] = dev;
3029 mdev->upper[port] = NULL;
3031 netif_carrier_off(dev);
3032 mlx4_en_set_default_moderation(priv);
3034 en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
3035 en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3037 mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3039 /* Configure port */
3040 mlx4_en_calc_rx_buf(dev);
3041 err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3042 priv->rx_skb_size + ETH_FCS_LEN,
3043 prof->tx_pause, prof->tx_ppp,
3044 prof->rx_pause, prof->rx_ppp);
3046 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3051 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3052 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3054 en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3061 en_warn(priv, "Initializing port\n");
3062 err = mlx4_INIT_PORT(mdev->dev, priv->port);
3064 en_err(priv, "Failed Initializing port\n");
3067 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3069 /* Initialize time stamp mechanism */
3070 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3071 mlx4_en_init_timestamp(mdev);
3073 queue_delayed_work(mdev->workqueue, &priv->service_task,
3074 SERVICE_TASK_DELAY);
3076 mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3077 mdev->profile.prof[priv->port].rx_ppp,
3078 mdev->profile.prof[priv->port].rx_pause,
3079 mdev->profile.prof[priv->port].tx_ppp,
3080 mdev->profile.prof[priv->port].tx_pause);
3082 err = register_netdev(dev);
3084 en_err(priv, "Netdev registration failed for port %d\n", port);
3088 priv->registered = 1;
3089 devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3095 mlx4_en_destroy_netdev(dev);
3099 int mlx4_en_reset_config(struct net_device *dev,
3100 struct hwtstamp_config ts_config,
3101 netdev_features_t features)
3103 struct mlx4_en_priv *priv = netdev_priv(dev);
3104 struct mlx4_en_dev *mdev = priv->mdev;
3108 if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3109 priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3110 !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3111 !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3112 return 0; /* Nothing to change */
3114 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3115 (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3116 (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3117 en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3121 mutex_lock(&mdev->state_lock);
3122 if (priv->port_up) {
3124 mlx4_en_stop_port(dev, 1);
3127 mlx4_en_free_resources(priv);
3129 en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
3130 ts_config.rx_filter, !!(features & NETIF_F_HW_VLAN_CTAG_RX));
3132 priv->hwtstamp_config.tx_type = ts_config.tx_type;
3133 priv->hwtstamp_config.rx_filter = ts_config.rx_filter;
3135 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3136 if (features & NETIF_F_HW_VLAN_CTAG_RX)
3137 dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3139 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3140 } else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3141 /* RX time-stamping is OFF, update the RX vlan offload
3142 * to the latest wanted state
3144 if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3145 dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3147 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3150 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3151 if (features & NETIF_F_RXFCS)
3152 dev->features |= NETIF_F_RXFCS;
3154 dev->features &= ~NETIF_F_RXFCS;
3157 /* RX vlan offload and RX time-stamping can't co-exist !
3158 * Regardless of the caller's choice,
3159 * Turn Off RX vlan offload in case of time-stamping is ON
3161 if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3162 if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3163 en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3164 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3167 err = mlx4_en_alloc_resources(priv);
3169 en_err(priv, "Failed reallocating port resources\n");
3173 err = mlx4_en_start_port(dev);
3175 en_err(priv, "Failed starting port\n");
3179 mutex_unlock(&mdev->state_lock);
3180 netdev_features_change(dev);