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");
414 err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
416 en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
419 mutex_unlock(&mdev->state_lock);
423 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
424 __be16 proto, u16 vid)
426 struct mlx4_en_priv *priv = netdev_priv(dev);
427 struct mlx4_en_dev *mdev = priv->mdev;
430 en_dbg(HW, priv, "Killing VID:%d\n", vid);
432 clear_bit(vid, priv->active_vlans);
434 /* Remove VID from port VLAN filter */
435 mutex_lock(&mdev->state_lock);
436 mlx4_unregister_vlan(mdev->dev, priv->port, vid);
438 if (mdev->device_up && priv->port_up) {
439 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
441 en_err(priv, "Failed configuring VLAN filter\n");
443 mutex_unlock(&mdev->state_lock);
448 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
451 for (i = ETH_ALEN - 1; i >= 0; --i) {
452 dst_mac[i] = src_mac & 0xff;
455 memset(&dst_mac[ETH_ALEN], 0, 2);
459 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
460 int qpn, u64 *reg_id)
464 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
465 priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
466 return 0; /* do nothing */
468 err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
469 MLX4_DOMAIN_NIC, reg_id);
471 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
474 en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
479 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
480 unsigned char *mac, int *qpn, u64 *reg_id)
482 struct mlx4_en_dev *mdev = priv->mdev;
483 struct mlx4_dev *dev = mdev->dev;
486 switch (dev->caps.steering_mode) {
487 case MLX4_STEERING_MODE_B0: {
492 memcpy(&gid[10], mac, ETH_ALEN);
495 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
498 case MLX4_STEERING_MODE_DEVICE_MANAGED: {
499 struct mlx4_spec_list spec_eth = { {NULL} };
500 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
502 struct mlx4_net_trans_rule rule = {
503 .queue_mode = MLX4_NET_TRANS_Q_FIFO,
506 .promisc_mode = MLX4_FS_REGULAR,
507 .priority = MLX4_DOMAIN_NIC,
510 rule.port = priv->port;
512 INIT_LIST_HEAD(&rule.list);
514 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
515 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
516 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
517 list_add_tail(&spec_eth.list, &rule.list);
519 err = mlx4_flow_attach(dev, &rule, reg_id);
526 en_warn(priv, "Failed Attaching Unicast\n");
531 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
532 unsigned char *mac, int qpn, u64 reg_id)
534 struct mlx4_en_dev *mdev = priv->mdev;
535 struct mlx4_dev *dev = mdev->dev;
537 switch (dev->caps.steering_mode) {
538 case MLX4_STEERING_MODE_B0: {
543 memcpy(&gid[10], mac, ETH_ALEN);
546 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
549 case MLX4_STEERING_MODE_DEVICE_MANAGED: {
550 mlx4_flow_detach(dev, reg_id);
554 en_err(priv, "Invalid steering mode.\n");
558 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
560 struct mlx4_en_dev *mdev = priv->mdev;
561 struct mlx4_dev *dev = mdev->dev;
564 int *qpn = &priv->base_qpn;
565 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
567 en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
568 priv->dev->dev_addr);
569 index = mlx4_register_mac(dev, priv->port, mac);
572 en_err(priv, "Failed adding MAC: %pM\n",
573 priv->dev->dev_addr);
577 if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
578 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
579 *qpn = base_qpn + index;
583 err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP);
584 en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
586 en_err(priv, "Failed to reserve qp for mac registration\n");
587 mlx4_unregister_mac(dev, priv->port, mac);
594 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
596 struct mlx4_en_dev *mdev = priv->mdev;
597 struct mlx4_dev *dev = mdev->dev;
598 int qpn = priv->base_qpn;
600 if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
601 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
602 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
603 priv->dev->dev_addr);
604 mlx4_unregister_mac(dev, priv->port, mac);
606 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
608 mlx4_qp_release_range(dev, qpn, 1);
609 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
613 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
614 unsigned char *new_mac, unsigned char *prev_mac)
616 struct mlx4_en_dev *mdev = priv->mdev;
617 struct mlx4_dev *dev = mdev->dev;
619 u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
621 if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
622 struct hlist_head *bucket;
623 unsigned int mac_hash;
624 struct mlx4_mac_entry *entry;
625 struct hlist_node *tmp;
626 u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
628 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
629 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
630 if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
631 mlx4_en_uc_steer_release(priv, entry->mac,
633 mlx4_unregister_mac(dev, priv->port,
635 hlist_del_rcu(&entry->hlist);
637 memcpy(entry->mac, new_mac, ETH_ALEN);
639 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
640 hlist_add_head_rcu(&entry->hlist,
641 &priv->mac_hash[mac_hash]);
642 mlx4_register_mac(dev, priv->port, new_mac_u64);
643 err = mlx4_en_uc_steer_add(priv, new_mac,
648 if (priv->tunnel_reg_id) {
649 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
650 priv->tunnel_reg_id = 0;
652 err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
653 &priv->tunnel_reg_id);
660 return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
663 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
664 unsigned char new_mac[ETH_ALEN + 2])
669 /* Remove old MAC and insert the new one */
670 err = mlx4_en_replace_mac(priv, priv->base_qpn,
671 new_mac, priv->current_mac);
673 en_err(priv, "Failed changing HW MAC address\n");
675 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
678 memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
683 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
685 struct mlx4_en_priv *priv = netdev_priv(dev);
686 struct mlx4_en_dev *mdev = priv->mdev;
687 struct sockaddr *saddr = addr;
688 unsigned char new_mac[ETH_ALEN + 2];
691 if (!is_valid_ether_addr(saddr->sa_data))
692 return -EADDRNOTAVAIL;
694 mutex_lock(&mdev->state_lock);
695 memcpy(new_mac, saddr->sa_data, ETH_ALEN);
696 err = mlx4_en_do_set_mac(priv, new_mac);
698 memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
699 mutex_unlock(&mdev->state_lock);
704 static void mlx4_en_clear_list(struct net_device *dev)
706 struct mlx4_en_priv *priv = netdev_priv(dev);
707 struct mlx4_en_mc_list *tmp, *mc_to_del;
709 list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
710 list_del(&mc_to_del->list);
715 static void mlx4_en_cache_mclist(struct net_device *dev)
717 struct mlx4_en_priv *priv = netdev_priv(dev);
718 struct netdev_hw_addr *ha;
719 struct mlx4_en_mc_list *tmp;
721 mlx4_en_clear_list(dev);
722 netdev_for_each_mc_addr(ha, dev) {
723 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
725 mlx4_en_clear_list(dev);
728 memcpy(tmp->addr, ha->addr, ETH_ALEN);
729 list_add_tail(&tmp->list, &priv->mc_list);
733 static void update_mclist_flags(struct mlx4_en_priv *priv,
734 struct list_head *dst,
735 struct list_head *src)
737 struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
740 /* Find all the entries that should be removed from dst,
741 * These are the entries that are not found in src
743 list_for_each_entry(dst_tmp, dst, list) {
745 list_for_each_entry(src_tmp, src, list) {
746 if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
752 dst_tmp->action = MCLIST_REM;
755 /* Add entries that exist in src but not in dst
756 * mark them as need to add
758 list_for_each_entry(src_tmp, src, list) {
760 list_for_each_entry(dst_tmp, dst, list) {
761 if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
762 dst_tmp->action = MCLIST_NONE;
768 new_mc = kmemdup(src_tmp,
769 sizeof(struct mlx4_en_mc_list),
774 new_mc->action = MCLIST_ADD;
775 list_add_tail(&new_mc->list, dst);
780 static void mlx4_en_set_rx_mode(struct net_device *dev)
782 struct mlx4_en_priv *priv = netdev_priv(dev);
787 queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
790 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
791 struct mlx4_en_dev *mdev)
795 if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
796 if (netif_msg_rx_status(priv))
797 en_warn(priv, "Entering promiscuous mode\n");
798 priv->flags |= MLX4_EN_FLAG_PROMISC;
800 /* Enable promiscouos mode */
801 switch (mdev->dev->caps.steering_mode) {
802 case MLX4_STEERING_MODE_DEVICE_MANAGED:
803 err = mlx4_flow_steer_promisc_add(mdev->dev,
806 MLX4_FS_ALL_DEFAULT);
808 en_err(priv, "Failed enabling promiscuous mode\n");
809 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
812 case MLX4_STEERING_MODE_B0:
813 err = mlx4_unicast_promisc_add(mdev->dev,
817 en_err(priv, "Failed enabling unicast promiscuous mode\n");
819 /* Add the default qp number as multicast
822 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
823 err = mlx4_multicast_promisc_add(mdev->dev,
827 en_err(priv, "Failed enabling multicast promiscuous mode\n");
828 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
832 case MLX4_STEERING_MODE_A0:
833 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
838 en_err(priv, "Failed enabling promiscuous mode\n");
842 /* Disable port multicast filter (unconditionally) */
843 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
844 0, MLX4_MCAST_DISABLE);
846 en_err(priv, "Failed disabling multicast filter\n");
850 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
851 struct mlx4_en_dev *mdev)
855 if (netif_msg_rx_status(priv))
856 en_warn(priv, "Leaving promiscuous mode\n");
857 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
859 /* Disable promiscouos mode */
860 switch (mdev->dev->caps.steering_mode) {
861 case MLX4_STEERING_MODE_DEVICE_MANAGED:
862 err = mlx4_flow_steer_promisc_remove(mdev->dev,
864 MLX4_FS_ALL_DEFAULT);
866 en_err(priv, "Failed disabling promiscuous mode\n");
867 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
870 case MLX4_STEERING_MODE_B0:
871 err = mlx4_unicast_promisc_remove(mdev->dev,
875 en_err(priv, "Failed disabling unicast promiscuous mode\n");
876 /* Disable Multicast promisc */
877 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
878 err = mlx4_multicast_promisc_remove(mdev->dev,
882 en_err(priv, "Failed disabling multicast promiscuous mode\n");
883 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
887 case MLX4_STEERING_MODE_A0:
888 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
892 en_err(priv, "Failed disabling promiscuous mode\n");
897 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
898 struct net_device *dev,
899 struct mlx4_en_dev *mdev)
901 struct mlx4_en_mc_list *mclist, *tmp;
903 u8 mc_list[16] = {0};
906 /* Enable/disable the multicast filter according to IFF_ALLMULTI */
907 if (dev->flags & IFF_ALLMULTI) {
908 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
909 0, MLX4_MCAST_DISABLE);
911 en_err(priv, "Failed disabling multicast filter\n");
913 /* Add the default qp number as multicast promisc */
914 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
915 switch (mdev->dev->caps.steering_mode) {
916 case MLX4_STEERING_MODE_DEVICE_MANAGED:
917 err = mlx4_flow_steer_promisc_add(mdev->dev,
923 case MLX4_STEERING_MODE_B0:
924 err = mlx4_multicast_promisc_add(mdev->dev,
929 case MLX4_STEERING_MODE_A0:
933 en_err(priv, "Failed entering multicast promisc mode\n");
934 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
937 /* Disable Multicast promisc */
938 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
939 switch (mdev->dev->caps.steering_mode) {
940 case MLX4_STEERING_MODE_DEVICE_MANAGED:
941 err = mlx4_flow_steer_promisc_remove(mdev->dev,
946 case MLX4_STEERING_MODE_B0:
947 err = mlx4_multicast_promisc_remove(mdev->dev,
952 case MLX4_STEERING_MODE_A0:
956 en_err(priv, "Failed disabling multicast promiscuous mode\n");
957 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
960 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
961 0, MLX4_MCAST_DISABLE);
963 en_err(priv, "Failed disabling multicast filter\n");
965 /* Flush mcast filter and init it with broadcast address */
966 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
967 1, MLX4_MCAST_CONFIG);
969 /* Update multicast list - we cache all addresses so they won't
970 * change while HW is updated holding the command semaphor */
971 netif_addr_lock_bh(dev);
972 mlx4_en_cache_mclist(dev);
973 netif_addr_unlock_bh(dev);
974 list_for_each_entry(mclist, &priv->mc_list, list) {
975 mcast_addr = mlx4_mac_to_u64(mclist->addr);
976 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
977 mcast_addr, 0, MLX4_MCAST_CONFIG);
979 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
980 0, MLX4_MCAST_ENABLE);
982 en_err(priv, "Failed enabling multicast filter\n");
984 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
985 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
986 if (mclist->action == MCLIST_REM) {
987 /* detach this address and delete from list */
988 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
989 mc_list[5] = priv->port;
990 err = mlx4_multicast_detach(mdev->dev,
991 &priv->rss_map.indir_qp,
996 en_err(priv, "Fail to detach multicast address\n");
998 if (mclist->tunnel_reg_id) {
999 err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1001 en_err(priv, "Failed to detach multicast address\n");
1004 /* remove from list */
1005 list_del(&mclist->list);
1007 } else if (mclist->action == MCLIST_ADD) {
1008 /* attach the address */
1009 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1010 /* needed for B0 steering support */
1011 mc_list[5] = priv->port;
1012 err = mlx4_multicast_attach(mdev->dev,
1013 &priv->rss_map.indir_qp,
1019 en_err(priv, "Fail to attach multicast address\n");
1021 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1022 &mclist->tunnel_reg_id);
1024 en_err(priv, "Failed to attach multicast address\n");
1030 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1031 struct net_device *dev,
1032 struct mlx4_en_dev *mdev)
1034 struct netdev_hw_addr *ha;
1035 struct mlx4_mac_entry *entry;
1036 struct hlist_node *tmp;
1040 struct hlist_head *bucket;
1045 /* Note that we do not need to protect our mac_hash traversal with rcu,
1046 * since all modification code is protected by mdev->state_lock
1049 /* find what to remove */
1050 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1051 bucket = &priv->mac_hash[i];
1052 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1054 netdev_for_each_uc_addr(ha, dev) {
1055 if (ether_addr_equal_64bits(entry->mac,
1062 /* MAC address of the port is not in uc list */
1063 if (ether_addr_equal_64bits(entry->mac,
1068 mac = mlx4_mac_to_u64(entry->mac);
1069 mlx4_en_uc_steer_release(priv, entry->mac,
1072 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1074 hlist_del_rcu(&entry->hlist);
1075 kfree_rcu(entry, rcu);
1076 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1077 entry->mac, priv->port);
1083 /* if we didn't remove anything, there is no use in trying to add
1084 * again once we are in a forced promisc mode state
1086 if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1089 prev_flags = priv->flags;
1090 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1092 /* find what to add */
1093 netdev_for_each_uc_addr(ha, dev) {
1095 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1096 hlist_for_each_entry(entry, bucket, hlist) {
1097 if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1104 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1106 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1107 ha->addr, priv->port);
1108 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1111 mac = mlx4_mac_to_u64(ha->addr);
1112 memcpy(entry->mac, ha->addr, ETH_ALEN);
1113 err = mlx4_register_mac(mdev->dev, priv->port, mac);
1115 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1116 ha->addr, priv->port, err);
1118 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1121 err = mlx4_en_uc_steer_add(priv, ha->addr,
1125 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1126 ha->addr, priv->port, err);
1127 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1129 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1132 unsigned int mac_hash;
1133 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1134 ha->addr, priv->port);
1135 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1136 bucket = &priv->mac_hash[mac_hash];
1137 hlist_add_head_rcu(&entry->hlist, bucket);
1142 if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1143 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1145 } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1146 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1151 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1153 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1155 struct mlx4_en_dev *mdev = priv->mdev;
1156 struct net_device *dev = priv->dev;
1158 mutex_lock(&mdev->state_lock);
1159 if (!mdev->device_up) {
1160 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1163 if (!priv->port_up) {
1164 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1168 if (!netif_carrier_ok(dev)) {
1169 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1170 if (priv->port_state.link_state) {
1171 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1172 netif_carrier_on(dev);
1173 en_dbg(LINK, priv, "Link Up\n");
1178 if (dev->priv_flags & IFF_UNICAST_FLT)
1179 mlx4_en_do_uc_filter(priv, dev, mdev);
1181 /* Promsicuous mode: disable all filters */
1182 if ((dev->flags & IFF_PROMISC) ||
1183 (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1184 mlx4_en_set_promisc_mode(priv, mdev);
1188 /* Not in promiscuous mode */
1189 if (priv->flags & MLX4_EN_FLAG_PROMISC)
1190 mlx4_en_clear_promisc_mode(priv, mdev);
1192 mlx4_en_do_multicast(priv, dev, mdev);
1194 mutex_unlock(&mdev->state_lock);
1197 #ifdef CONFIG_NET_POLL_CONTROLLER
1198 static void mlx4_en_netpoll(struct net_device *dev)
1200 struct mlx4_en_priv *priv = netdev_priv(dev);
1201 struct mlx4_en_cq *cq;
1204 for (i = 0; i < priv->rx_ring_num; i++) {
1205 cq = priv->rx_cq[i];
1206 napi_schedule(&cq->napi);
1211 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1215 int *qpn = &priv->base_qpn;
1216 struct mlx4_mac_entry *entry;
1218 err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, ®_id);
1222 err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1223 &priv->tunnel_reg_id);
1227 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1233 memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1234 memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1235 entry->reg_id = reg_id;
1236 hlist_add_head_rcu(&entry->hlist,
1237 &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1242 if (priv->tunnel_reg_id)
1243 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1246 mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1250 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1254 int qpn = priv->base_qpn;
1255 struct hlist_head *bucket;
1256 struct hlist_node *tmp;
1257 struct mlx4_mac_entry *entry;
1259 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1260 bucket = &priv->mac_hash[i];
1261 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1262 mac = mlx4_mac_to_u64(entry->mac);
1263 en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1265 mlx4_en_uc_steer_release(priv, entry->mac,
1266 qpn, entry->reg_id);
1268 mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1269 hlist_del_rcu(&entry->hlist);
1270 kfree_rcu(entry, rcu);
1274 if (priv->tunnel_reg_id) {
1275 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1276 priv->tunnel_reg_id = 0;
1280 static void mlx4_en_tx_timeout(struct net_device *dev)
1282 struct mlx4_en_priv *priv = netdev_priv(dev);
1283 struct mlx4_en_dev *mdev = priv->mdev;
1286 if (netif_msg_timer(priv))
1287 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1289 for (i = 0; i < priv->tx_ring_num; i++) {
1290 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1292 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1293 i, priv->tx_ring[i]->qpn, priv->tx_ring[i]->cqn,
1294 priv->tx_ring[i]->cons, priv->tx_ring[i]->prod);
1297 priv->port_stats.tx_timeout++;
1298 en_dbg(DRV, priv, "Scheduling watchdog\n");
1299 queue_work(mdev->workqueue, &priv->watchdog_task);
1303 static struct rtnl_link_stats64 *
1304 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1306 struct mlx4_en_priv *priv = netdev_priv(dev);
1308 spin_lock_bh(&priv->stats_lock);
1309 netdev_stats_to_stats64(stats, &dev->stats);
1310 spin_unlock_bh(&priv->stats_lock);
1315 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1317 struct mlx4_en_cq *cq;
1320 /* If we haven't received a specific coalescing setting
1321 * (module param), we set the moderation parameters as follows:
1322 * - moder_cnt is set to the number of mtu sized packets to
1323 * satisfy our coalescing target.
1324 * - moder_time is set to a fixed value.
1326 priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1327 priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1328 priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1329 priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1330 en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1331 priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1333 /* Setup cq moderation params */
1334 for (i = 0; i < priv->rx_ring_num; i++) {
1335 cq = priv->rx_cq[i];
1336 cq->moder_cnt = priv->rx_frames;
1337 cq->moder_time = priv->rx_usecs;
1338 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1339 priv->last_moder_packets[i] = 0;
1340 priv->last_moder_bytes[i] = 0;
1343 for (i = 0; i < priv->tx_ring_num; i++) {
1344 cq = priv->tx_cq[i];
1345 cq->moder_cnt = priv->tx_frames;
1346 cq->moder_time = priv->tx_usecs;
1349 /* Reset auto-moderation params */
1350 priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1351 priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1352 priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1353 priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1354 priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1355 priv->adaptive_rx_coal = 1;
1356 priv->last_moder_jiffies = 0;
1357 priv->last_moder_tx_packets = 0;
1360 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1362 unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1363 struct mlx4_en_cq *cq;
1364 unsigned long packets;
1366 unsigned long avg_pkt_size;
1367 unsigned long rx_packets;
1368 unsigned long rx_bytes;
1369 unsigned long rx_pkt_diff;
1373 if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1376 for (ring = 0; ring < priv->rx_ring_num; ring++) {
1377 spin_lock_bh(&priv->stats_lock);
1378 rx_packets = priv->rx_ring[ring]->packets;
1379 rx_bytes = priv->rx_ring[ring]->bytes;
1380 spin_unlock_bh(&priv->stats_lock);
1382 rx_pkt_diff = ((unsigned long) (rx_packets -
1383 priv->last_moder_packets[ring]));
1384 packets = rx_pkt_diff;
1385 rate = packets * HZ / period;
1386 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1387 priv->last_moder_bytes[ring])) / packets : 0;
1389 /* Apply auto-moderation only when packet rate
1390 * exceeds a rate that it matters */
1391 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1392 avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1393 if (rate < priv->pkt_rate_low)
1394 moder_time = priv->rx_usecs_low;
1395 else if (rate > priv->pkt_rate_high)
1396 moder_time = priv->rx_usecs_high;
1398 moder_time = (rate - priv->pkt_rate_low) *
1399 (priv->rx_usecs_high - priv->rx_usecs_low) /
1400 (priv->pkt_rate_high - priv->pkt_rate_low) +
1403 moder_time = priv->rx_usecs_low;
1406 if (moder_time != priv->last_moder_time[ring]) {
1407 priv->last_moder_time[ring] = moder_time;
1408 cq = priv->rx_cq[ring];
1409 cq->moder_time = moder_time;
1410 cq->moder_cnt = priv->rx_frames;
1411 err = mlx4_en_set_cq_moder(priv, cq);
1413 en_err(priv, "Failed modifying moderation for cq:%d\n",
1416 priv->last_moder_packets[ring] = rx_packets;
1417 priv->last_moder_bytes[ring] = rx_bytes;
1420 priv->last_moder_jiffies = jiffies;
1423 static void mlx4_en_do_get_stats(struct work_struct *work)
1425 struct delayed_work *delay = to_delayed_work(work);
1426 struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1428 struct mlx4_en_dev *mdev = priv->mdev;
1431 mutex_lock(&mdev->state_lock);
1432 if (mdev->device_up) {
1433 if (priv->port_up) {
1434 err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1436 en_dbg(HW, priv, "Could not update stats\n");
1438 mlx4_en_auto_moderation(priv);
1441 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1443 if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1444 mlx4_en_do_set_mac(priv, priv->current_mac);
1445 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1447 mutex_unlock(&mdev->state_lock);
1450 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1453 static void mlx4_en_service_task(struct work_struct *work)
1455 struct delayed_work *delay = to_delayed_work(work);
1456 struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1458 struct mlx4_en_dev *mdev = priv->mdev;
1460 mutex_lock(&mdev->state_lock);
1461 if (mdev->device_up) {
1462 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1463 mlx4_en_ptp_overflow_check(mdev);
1465 mlx4_en_recover_from_oom(priv);
1466 queue_delayed_work(mdev->workqueue, &priv->service_task,
1467 SERVICE_TASK_DELAY);
1469 mutex_unlock(&mdev->state_lock);
1472 static void mlx4_en_linkstate(struct work_struct *work)
1474 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1476 struct mlx4_en_dev *mdev = priv->mdev;
1477 int linkstate = priv->link_state;
1479 mutex_lock(&mdev->state_lock);
1480 /* If observable port state changed set carrier state and
1481 * report to system log */
1482 if (priv->last_link_state != linkstate) {
1483 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1484 en_info(priv, "Link Down\n");
1485 netif_carrier_off(priv->dev);
1487 en_info(priv, "Link Up\n");
1488 netif_carrier_on(priv->dev);
1491 priv->last_link_state = linkstate;
1492 mutex_unlock(&mdev->state_lock);
1495 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1497 struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1498 int numa_node = priv->mdev->dev->numa_node;
1500 if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1503 cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1504 ring->affinity_mask);
1508 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1510 free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1513 int mlx4_en_start_port(struct net_device *dev)
1515 struct mlx4_en_priv *priv = netdev_priv(dev);
1516 struct mlx4_en_dev *mdev = priv->mdev;
1517 struct mlx4_en_cq *cq;
1518 struct mlx4_en_tx_ring *tx_ring;
1524 u8 mc_list[16] = {0};
1526 if (priv->port_up) {
1527 en_dbg(DRV, priv, "start port called while port already up\n");
1531 INIT_LIST_HEAD(&priv->mc_list);
1532 INIT_LIST_HEAD(&priv->curr_list);
1533 INIT_LIST_HEAD(&priv->ethtool_list);
1534 memset(&priv->ethtool_rules[0], 0,
1535 sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1537 /* Calculate Rx buf size */
1538 dev->mtu = min(dev->mtu, priv->max_mtu);
1539 mlx4_en_calc_rx_buf(dev);
1540 en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1542 /* Configure rx cq's and rings */
1543 err = mlx4_en_activate_rx_rings(priv);
1545 en_err(priv, "Failed to activate RX rings\n");
1548 for (i = 0; i < priv->rx_ring_num; i++) {
1549 cq = priv->rx_cq[i];
1551 err = mlx4_en_init_affinity_hint(priv, i);
1553 en_err(priv, "Failed preparing IRQ affinity hint\n");
1557 err = mlx4_en_activate_cq(priv, cq, i);
1559 en_err(priv, "Failed activating Rx CQ\n");
1560 mlx4_en_free_affinity_hint(priv, i);
1564 for (j = 0; j < cq->size; j++) {
1565 struct mlx4_cqe *cqe = NULL;
1567 cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1569 cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1572 err = mlx4_en_set_cq_moder(priv, cq);
1574 en_err(priv, "Failed setting cq moderation parameters\n");
1575 mlx4_en_deactivate_cq(priv, cq);
1576 mlx4_en_free_affinity_hint(priv, i);
1579 mlx4_en_arm_cq(priv, cq);
1580 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1585 en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1586 err = mlx4_en_get_qp(priv);
1588 en_err(priv, "Failed getting eth qp\n");
1591 mdev->mac_removed[priv->port] = 0;
1593 priv->counter_index =
1594 mlx4_get_default_counter_index(mdev->dev, priv->port);
1596 err = mlx4_en_config_rss_steer(priv);
1598 en_err(priv, "Failed configuring rss steering\n");
1602 err = mlx4_en_create_drop_qp(priv);
1606 /* Configure tx cq's and rings */
1607 for (i = 0; i < priv->tx_ring_num; i++) {
1609 cq = priv->tx_cq[i];
1610 err = mlx4_en_activate_cq(priv, cq, i);
1612 en_err(priv, "Failed allocating Tx CQ\n");
1615 err = mlx4_en_set_cq_moder(priv, cq);
1617 en_err(priv, "Failed setting cq moderation parameters\n");
1618 mlx4_en_deactivate_cq(priv, cq);
1621 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1622 cq->buf->wqe_index = cpu_to_be16(0xffff);
1624 /* Configure ring */
1625 tx_ring = priv->tx_ring[i];
1626 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1627 i / priv->num_tx_rings_p_up);
1629 en_err(priv, "Failed allocating Tx ring\n");
1630 mlx4_en_deactivate_cq(priv, cq);
1633 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1635 /* Arm CQ for TX completions */
1636 mlx4_en_arm_cq(priv, cq);
1638 /* Set initial ownership of all Tx TXBBs to SW (1) */
1639 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1640 *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1644 /* Configure port */
1645 err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1646 priv->rx_skb_size + ETH_FCS_LEN,
1647 priv->prof->tx_pause,
1649 priv->prof->rx_pause,
1650 priv->prof->rx_ppp);
1652 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1656 /* Set default qp number */
1657 err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1659 en_err(priv, "Failed setting default qp numbers\n");
1663 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1664 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1666 en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1673 en_dbg(HW, priv, "Initializing port\n");
1674 err = mlx4_INIT_PORT(mdev->dev, priv->port);
1676 en_err(priv, "Failed Initializing port\n");
1680 /* Set Unicast and VXLAN steering rules */
1681 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1682 mlx4_en_set_rss_steer_rules(priv))
1683 mlx4_warn(mdev, "Failed setting steering rules\n");
1685 /* Attach rx QP to bradcast address */
1686 eth_broadcast_addr(&mc_list[10]);
1687 mc_list[5] = priv->port; /* needed for B0 steering support */
1688 if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1689 priv->port, 0, MLX4_PROT_ETH,
1690 &priv->broadcast_id))
1691 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1693 /* Must redo promiscuous mode setup. */
1694 priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1696 /* Schedule multicast task to populate multicast list */
1697 queue_work(mdev->workqueue, &priv->rx_mode_task);
1699 #ifdef CONFIG_MLX4_EN_VXLAN
1700 if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1701 vxlan_get_rx_port(dev);
1703 priv->port_up = true;
1704 netif_tx_start_all_queues(dev);
1705 netif_device_attach(dev);
1710 while (tx_index--) {
1711 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1712 mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1714 mlx4_en_destroy_drop_qp(priv);
1716 mlx4_en_release_rss_steer(priv);
1718 mlx4_en_put_qp(priv);
1720 while (rx_index--) {
1721 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1722 mlx4_en_free_affinity_hint(priv, rx_index);
1724 for (i = 0; i < priv->rx_ring_num; i++)
1725 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1727 return err; /* need to close devices */
1731 void mlx4_en_stop_port(struct net_device *dev, int detach)
1733 struct mlx4_en_priv *priv = netdev_priv(dev);
1734 struct mlx4_en_dev *mdev = priv->mdev;
1735 struct mlx4_en_mc_list *mclist, *tmp;
1736 struct ethtool_flow_id *flow, *tmp_flow;
1738 u8 mc_list[16] = {0};
1740 if (!priv->port_up) {
1741 en_dbg(DRV, priv, "stop port called while port already down\n");
1746 mlx4_CLOSE_PORT(mdev->dev, priv->port);
1748 /* Synchronize with tx routine */
1749 netif_tx_lock_bh(dev);
1751 netif_device_detach(dev);
1752 netif_tx_stop_all_queues(dev);
1753 netif_tx_unlock_bh(dev);
1755 netif_tx_disable(dev);
1757 /* Set port as not active */
1758 priv->port_up = false;
1759 priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1761 /* Promsicuous mode */
1762 if (mdev->dev->caps.steering_mode ==
1763 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1764 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1765 MLX4_EN_FLAG_MC_PROMISC);
1766 mlx4_flow_steer_promisc_remove(mdev->dev,
1768 MLX4_FS_ALL_DEFAULT);
1769 mlx4_flow_steer_promisc_remove(mdev->dev,
1771 MLX4_FS_MC_DEFAULT);
1772 } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1773 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1775 /* Disable promiscouos mode */
1776 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1779 /* Disable Multicast promisc */
1780 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1781 mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1783 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1787 /* Detach All multicasts */
1788 eth_broadcast_addr(&mc_list[10]);
1789 mc_list[5] = priv->port; /* needed for B0 steering support */
1790 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1791 MLX4_PROT_ETH, priv->broadcast_id);
1792 list_for_each_entry(mclist, &priv->curr_list, list) {
1793 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1794 mc_list[5] = priv->port;
1795 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1796 mc_list, MLX4_PROT_ETH, mclist->reg_id);
1797 if (mclist->tunnel_reg_id)
1798 mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1800 mlx4_en_clear_list(dev);
1801 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1802 list_del(&mclist->list);
1806 /* Flush multicast filter */
1807 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1809 /* Remove flow steering rules for the port*/
1810 if (mdev->dev->caps.steering_mode ==
1811 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1813 list_for_each_entry_safe(flow, tmp_flow,
1814 &priv->ethtool_list, list) {
1815 mlx4_flow_detach(mdev->dev, flow->id);
1816 list_del(&flow->list);
1820 mlx4_en_destroy_drop_qp(priv);
1823 for (i = 0; i < priv->tx_ring_num; i++) {
1824 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1825 mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1829 for (i = 0; i < priv->tx_ring_num; i++)
1830 mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1832 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1833 mlx4_en_delete_rss_steer_rules(priv);
1836 mlx4_en_release_rss_steer(priv);
1838 /* Unregister Mac address for the port */
1839 mlx4_en_put_qp(priv);
1840 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1841 mdev->mac_removed[priv->port] = 1;
1844 for (i = 0; i < priv->rx_ring_num; i++) {
1845 struct mlx4_en_cq *cq = priv->rx_cq[i];
1847 napi_synchronize(&cq->napi);
1848 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1849 mlx4_en_deactivate_cq(priv, cq);
1851 mlx4_en_free_affinity_hint(priv, i);
1855 static void mlx4_en_restart(struct work_struct *work)
1857 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1859 struct mlx4_en_dev *mdev = priv->mdev;
1860 struct net_device *dev = priv->dev;
1862 en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1865 mutex_lock(&mdev->state_lock);
1866 if (priv->port_up) {
1867 mlx4_en_stop_port(dev, 1);
1868 if (mlx4_en_start_port(dev))
1869 en_err(priv, "Failed restarting port %d\n", priv->port);
1871 mutex_unlock(&mdev->state_lock);
1875 static void mlx4_en_clear_stats(struct net_device *dev)
1877 struct mlx4_en_priv *priv = netdev_priv(dev);
1878 struct mlx4_en_dev *mdev = priv->mdev;
1881 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1882 en_dbg(HW, priv, "Failed dumping statistics\n");
1884 memset(&priv->pstats, 0, sizeof(priv->pstats));
1885 memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1886 memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1887 memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
1888 memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
1889 memset(&priv->rx_priority_flowstats, 0,
1890 sizeof(priv->rx_priority_flowstats));
1891 memset(&priv->tx_priority_flowstats, 0,
1892 sizeof(priv->tx_priority_flowstats));
1893 memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
1895 for (i = 0; i < priv->tx_ring_num; i++) {
1896 priv->tx_ring[i]->bytes = 0;
1897 priv->tx_ring[i]->packets = 0;
1898 priv->tx_ring[i]->tx_csum = 0;
1899 priv->tx_ring[i]->tx_dropped = 0;
1900 priv->tx_ring[i]->queue_stopped = 0;
1901 priv->tx_ring[i]->wake_queue = 0;
1902 priv->tx_ring[i]->tso_packets = 0;
1903 priv->tx_ring[i]->xmit_more = 0;
1905 for (i = 0; i < priv->rx_ring_num; i++) {
1906 priv->rx_ring[i]->bytes = 0;
1907 priv->rx_ring[i]->packets = 0;
1908 priv->rx_ring[i]->csum_ok = 0;
1909 priv->rx_ring[i]->csum_none = 0;
1910 priv->rx_ring[i]->csum_complete = 0;
1914 static int mlx4_en_open(struct net_device *dev)
1916 struct mlx4_en_priv *priv = netdev_priv(dev);
1917 struct mlx4_en_dev *mdev = priv->mdev;
1920 mutex_lock(&mdev->state_lock);
1922 if (!mdev->device_up) {
1923 en_err(priv, "Cannot open - device down/disabled\n");
1928 /* Reset HW statistics and SW counters */
1929 mlx4_en_clear_stats(dev);
1931 err = mlx4_en_start_port(dev);
1933 en_err(priv, "Failed starting port:%d\n", priv->port);
1936 mutex_unlock(&mdev->state_lock);
1941 static int mlx4_en_close(struct net_device *dev)
1943 struct mlx4_en_priv *priv = netdev_priv(dev);
1944 struct mlx4_en_dev *mdev = priv->mdev;
1946 en_dbg(IFDOWN, priv, "Close port called\n");
1948 mutex_lock(&mdev->state_lock);
1950 mlx4_en_stop_port(dev, 0);
1951 netif_carrier_off(dev);
1953 mutex_unlock(&mdev->state_lock);
1957 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1961 #ifdef CONFIG_RFS_ACCEL
1962 priv->dev->rx_cpu_rmap = NULL;
1965 for (i = 0; i < priv->tx_ring_num; i++) {
1966 if (priv->tx_ring && priv->tx_ring[i])
1967 mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1968 if (priv->tx_cq && priv->tx_cq[i])
1969 mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1972 for (i = 0; i < priv->rx_ring_num; i++) {
1973 if (priv->rx_ring[i])
1974 mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1975 priv->prof->rx_ring_size, priv->stride);
1977 mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1982 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1984 struct mlx4_en_port_profile *prof = priv->prof;
1988 /* Create tx Rings */
1989 for (i = 0; i < priv->tx_ring_num; i++) {
1990 node = cpu_to_node(i % num_online_cpus());
1991 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1992 prof->tx_ring_size, i, TX, node))
1995 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
1996 prof->tx_ring_size, TXBB_SIZE,
2001 /* Create rx Rings */
2002 for (i = 0; i < priv->rx_ring_num; i++) {
2003 node = cpu_to_node(i % num_online_cpus());
2004 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2005 prof->rx_ring_size, i, RX, node))
2008 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2009 prof->rx_ring_size, priv->stride,
2014 #ifdef CONFIG_RFS_ACCEL
2015 priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2021 en_err(priv, "Failed to allocate NIC resources\n");
2022 for (i = 0; i < priv->rx_ring_num; i++) {
2023 if (priv->rx_ring[i])
2024 mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2028 mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2030 for (i = 0; i < priv->tx_ring_num; i++) {
2031 if (priv->tx_ring[i])
2032 mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
2034 mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
2039 static void mlx4_en_shutdown(struct net_device *dev)
2042 netif_device_detach(dev);
2047 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2048 struct mlx4_en_priv *src,
2049 struct mlx4_en_port_profile *prof)
2051 memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2052 sizeof(dst->hwtstamp_config));
2053 dst->num_tx_rings_p_up = src->mdev->profile.num_tx_rings_p_up;
2054 dst->tx_ring_num = prof->tx_ring_num;
2055 dst->rx_ring_num = prof->rx_ring_num;
2056 dst->flags = prof->flags;
2057 dst->mdev = src->mdev;
2058 dst->port = src->port;
2059 dst->dev = src->dev;
2061 dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2062 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2064 dst->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2069 dst->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2072 kfree(dst->tx_ring);
2078 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2079 struct mlx4_en_priv *src)
2081 memcpy(dst->rx_ring, src->rx_ring,
2082 sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2083 memcpy(dst->rx_cq, src->rx_cq,
2084 sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2085 memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2086 sizeof(dst->hwtstamp_config));
2087 dst->tx_ring_num = src->tx_ring_num;
2088 dst->rx_ring_num = src->rx_ring_num;
2089 dst->tx_ring = src->tx_ring;
2090 dst->tx_cq = src->tx_cq;
2091 memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2094 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2095 struct mlx4_en_priv *tmp,
2096 struct mlx4_en_port_profile *prof)
2098 mlx4_en_copy_priv(tmp, priv, prof);
2100 if (mlx4_en_alloc_resources(tmp)) {
2102 "%s: Resource allocation failed, using previous configuration\n",
2104 kfree(tmp->tx_ring);
2111 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2112 struct mlx4_en_priv *tmp)
2114 mlx4_en_free_resources(priv);
2115 mlx4_en_update_priv(priv, tmp);
2118 void mlx4_en_destroy_netdev(struct net_device *dev)
2120 struct mlx4_en_priv *priv = netdev_priv(dev);
2121 struct mlx4_en_dev *mdev = priv->mdev;
2122 bool shutdown = mdev->dev->persist->interface_state &
2123 MLX4_INTERFACE_STATE_SHUTDOWN;
2125 en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2127 /* Unregister device - this will close the port if it was up */
2128 if (priv->registered) {
2129 devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2132 mlx4_en_shutdown(dev);
2134 unregister_netdev(dev);
2137 if (priv->allocated)
2138 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2140 cancel_delayed_work(&priv->stats_task);
2141 cancel_delayed_work(&priv->service_task);
2142 /* flush any pending task for this netdev */
2143 flush_workqueue(mdev->workqueue);
2145 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2146 mlx4_en_remove_timestamp(mdev);
2148 /* Detach the netdev so tasks would not attempt to access it */
2149 mutex_lock(&mdev->state_lock);
2150 mdev->pndev[priv->port] = NULL;
2151 mdev->upper[priv->port] = NULL;
2152 mutex_unlock(&mdev->state_lock);
2154 #ifdef CONFIG_RFS_ACCEL
2155 mlx4_en_cleanup_filters(priv);
2158 mlx4_en_free_resources(priv);
2160 kfree(priv->tx_ring);
2167 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2169 struct mlx4_en_priv *priv = netdev_priv(dev);
2170 struct mlx4_en_dev *mdev = priv->mdev;
2173 en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2176 if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
2177 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
2182 if (netif_running(dev)) {
2183 mutex_lock(&mdev->state_lock);
2184 if (!mdev->device_up) {
2185 /* NIC is probably restarting - let watchdog task reset
2187 en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2189 mlx4_en_stop_port(dev, 1);
2190 err = mlx4_en_start_port(dev);
2192 en_err(priv, "Failed restarting port:%d\n",
2194 queue_work(mdev->workqueue, &priv->watchdog_task);
2197 mutex_unlock(&mdev->state_lock);
2202 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2204 struct mlx4_en_priv *priv = netdev_priv(dev);
2205 struct mlx4_en_dev *mdev = priv->mdev;
2206 struct hwtstamp_config config;
2208 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2211 /* reserved for future extensions */
2215 /* device doesn't support time stamping */
2216 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2219 /* TX HW timestamp */
2220 switch (config.tx_type) {
2221 case HWTSTAMP_TX_OFF:
2222 case HWTSTAMP_TX_ON:
2228 /* RX HW timestamp */
2229 switch (config.rx_filter) {
2230 case HWTSTAMP_FILTER_NONE:
2232 case HWTSTAMP_FILTER_ALL:
2233 case HWTSTAMP_FILTER_SOME:
2234 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2235 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2236 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2237 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2238 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2239 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2240 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2241 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2242 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2243 case HWTSTAMP_FILTER_PTP_V2_EVENT:
2244 case HWTSTAMP_FILTER_PTP_V2_SYNC:
2245 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2246 config.rx_filter = HWTSTAMP_FILTER_ALL;
2252 if (mlx4_en_reset_config(dev, config, dev->features)) {
2253 config.tx_type = HWTSTAMP_TX_OFF;
2254 config.rx_filter = HWTSTAMP_FILTER_NONE;
2257 return copy_to_user(ifr->ifr_data, &config,
2258 sizeof(config)) ? -EFAULT : 0;
2261 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2263 struct mlx4_en_priv *priv = netdev_priv(dev);
2265 return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2266 sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2269 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2273 return mlx4_en_hwtstamp_set(dev, ifr);
2275 return mlx4_en_hwtstamp_get(dev, ifr);
2281 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2282 netdev_features_t features)
2284 struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2285 struct mlx4_en_dev *mdev = en_priv->mdev;
2287 /* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2288 * enable/disable make sure S-TAG flag is always in same state as
2291 if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2292 !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2293 features |= NETIF_F_HW_VLAN_STAG_RX;
2295 features &= ~NETIF_F_HW_VLAN_STAG_RX;
2300 static int mlx4_en_set_features(struct net_device *netdev,
2301 netdev_features_t features)
2303 struct mlx4_en_priv *priv = netdev_priv(netdev);
2307 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2308 en_info(priv, "Turn %s RX-FCS\n",
2309 (features & NETIF_F_RXFCS) ? "ON" : "OFF");
2313 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2314 u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2316 en_info(priv, "Turn %s RX-ALL\n",
2317 ignore_fcs_value ? "ON" : "OFF");
2318 ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2319 priv->port, ignore_fcs_value);
2324 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2325 en_info(priv, "Turn %s RX vlan strip offload\n",
2326 (features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2330 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2331 en_info(priv, "Turn %s TX vlan strip offload\n",
2332 (features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2334 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2335 en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2336 (features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2338 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2339 en_info(priv, "Turn %s loopback\n",
2340 (features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2341 mlx4_en_update_loopback_state(netdev, features);
2345 ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2354 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2356 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2357 struct mlx4_en_dev *mdev = en_priv->mdev;
2358 u64 mac_u64 = mlx4_mac_to_u64(mac);
2360 if (is_multicast_ether_addr(mac))
2363 return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2366 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos)
2368 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2369 struct mlx4_en_dev *mdev = en_priv->mdev;
2371 return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos);
2374 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2377 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2378 struct mlx4_en_dev *mdev = en_priv->mdev;
2380 return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2384 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2386 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2387 struct mlx4_en_dev *mdev = en_priv->mdev;
2389 return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2392 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2394 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2395 struct mlx4_en_dev *mdev = en_priv->mdev;
2397 return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2400 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2402 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2403 struct mlx4_en_dev *mdev = en_priv->mdev;
2405 return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2408 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2409 struct ifla_vf_stats *vf_stats)
2411 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2412 struct mlx4_en_dev *mdev = en_priv->mdev;
2414 return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2417 #define PORT_ID_BYTE_LEN 8
2418 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2419 struct netdev_phys_item_id *ppid)
2421 struct mlx4_en_priv *priv = netdev_priv(dev);
2422 struct mlx4_dev *mdev = priv->mdev->dev;
2424 u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2429 ppid->id_len = sizeof(phys_port_id);
2430 for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2431 ppid->id[i] = phys_port_id & 0xff;
2437 #ifdef CONFIG_MLX4_EN_VXLAN
2438 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2441 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2444 ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2448 ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2449 VXLAN_STEER_BY_OUTER_MAC, 1);
2452 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2457 priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2459 NETIF_F_TSO | NETIF_F_TSO6 |
2460 NETIF_F_GSO_UDP_TUNNEL |
2461 NETIF_F_GSO_UDP_TUNNEL_CSUM |
2462 NETIF_F_GSO_PARTIAL;
2465 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2468 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2470 /* unset offloads */
2471 priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2473 NETIF_F_TSO | NETIF_F_TSO6 |
2474 NETIF_F_GSO_UDP_TUNNEL |
2475 NETIF_F_GSO_UDP_TUNNEL_CSUM |
2476 NETIF_F_GSO_PARTIAL);
2478 ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2479 VXLAN_STEER_BY_OUTER_MAC, 0);
2481 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2483 priv->vxlan_port = 0;
2486 static void mlx4_en_add_vxlan_port(struct net_device *dev,
2487 sa_family_t sa_family, __be16 port)
2489 struct mlx4_en_priv *priv = netdev_priv(dev);
2490 __be16 current_port;
2492 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2495 if (sa_family == AF_INET6)
2498 current_port = priv->vxlan_port;
2499 if (current_port && current_port != port) {
2500 en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2501 ntohs(current_port), ntohs(port));
2505 priv->vxlan_port = port;
2506 queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2509 static void mlx4_en_del_vxlan_port(struct net_device *dev,
2510 sa_family_t sa_family, __be16 port)
2512 struct mlx4_en_priv *priv = netdev_priv(dev);
2513 __be16 current_port;
2515 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2518 if (sa_family == AF_INET6)
2521 current_port = priv->vxlan_port;
2522 if (current_port != port) {
2523 en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2527 queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2530 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2531 struct net_device *dev,
2532 netdev_features_t features)
2534 features = vlan_features_check(skb, features);
2535 features = vxlan_features_check(skb, features);
2537 /* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2538 * support inner IPv6 checksums and segmentation so we need to
2539 * strip that feature if this is an IPv6 encapsulated frame.
2541 if (skb->encapsulation &&
2542 (skb->ip_summed == CHECKSUM_PARTIAL)) {
2543 struct mlx4_en_priv *priv = netdev_priv(dev);
2545 if (!priv->vxlan_port ||
2546 (ip_hdr(skb)->version != 4) ||
2547 (udp_hdr(skb)->dest != priv->vxlan_port))
2548 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2555 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2557 struct mlx4_en_priv *priv = netdev_priv(dev);
2558 struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[queue_index];
2559 struct mlx4_update_qp_params params;
2562 if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2565 /* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2566 if (maxrate >> 12) {
2567 params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2568 params.rate_val = maxrate / 1000;
2569 } else if (maxrate) {
2570 params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2571 params.rate_val = maxrate;
2572 } else { /* zero serves to revoke the QP rate-limitation */
2573 params.rate_unit = 0;
2574 params.rate_val = 0;
2577 err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2582 static const struct net_device_ops mlx4_netdev_ops = {
2583 .ndo_open = mlx4_en_open,
2584 .ndo_stop = mlx4_en_close,
2585 .ndo_start_xmit = mlx4_en_xmit,
2586 .ndo_select_queue = mlx4_en_select_queue,
2587 .ndo_get_stats64 = mlx4_en_get_stats64,
2588 .ndo_set_rx_mode = mlx4_en_set_rx_mode,
2589 .ndo_set_mac_address = mlx4_en_set_mac,
2590 .ndo_validate_addr = eth_validate_addr,
2591 .ndo_change_mtu = mlx4_en_change_mtu,
2592 .ndo_do_ioctl = mlx4_en_ioctl,
2593 .ndo_tx_timeout = mlx4_en_tx_timeout,
2594 .ndo_vlan_rx_add_vid = mlx4_en_vlan_rx_add_vid,
2595 .ndo_vlan_rx_kill_vid = mlx4_en_vlan_rx_kill_vid,
2596 #ifdef CONFIG_NET_POLL_CONTROLLER
2597 .ndo_poll_controller = mlx4_en_netpoll,
2599 .ndo_set_features = mlx4_en_set_features,
2600 .ndo_fix_features = mlx4_en_fix_features,
2601 .ndo_setup_tc = __mlx4_en_setup_tc,
2602 #ifdef CONFIG_RFS_ACCEL
2603 .ndo_rx_flow_steer = mlx4_en_filter_rfs,
2605 .ndo_get_phys_port_id = mlx4_en_get_phys_port_id,
2606 #ifdef CONFIG_MLX4_EN_VXLAN
2607 .ndo_add_vxlan_port = mlx4_en_add_vxlan_port,
2608 .ndo_del_vxlan_port = mlx4_en_del_vxlan_port,
2609 .ndo_features_check = mlx4_en_features_check,
2611 .ndo_set_tx_maxrate = mlx4_en_set_tx_maxrate,
2614 static const struct net_device_ops mlx4_netdev_ops_master = {
2615 .ndo_open = mlx4_en_open,
2616 .ndo_stop = mlx4_en_close,
2617 .ndo_start_xmit = mlx4_en_xmit,
2618 .ndo_select_queue = mlx4_en_select_queue,
2619 .ndo_get_stats64 = mlx4_en_get_stats64,
2620 .ndo_set_rx_mode = mlx4_en_set_rx_mode,
2621 .ndo_set_mac_address = mlx4_en_set_mac,
2622 .ndo_validate_addr = eth_validate_addr,
2623 .ndo_change_mtu = mlx4_en_change_mtu,
2624 .ndo_tx_timeout = mlx4_en_tx_timeout,
2625 .ndo_vlan_rx_add_vid = mlx4_en_vlan_rx_add_vid,
2626 .ndo_vlan_rx_kill_vid = mlx4_en_vlan_rx_kill_vid,
2627 .ndo_set_vf_mac = mlx4_en_set_vf_mac,
2628 .ndo_set_vf_vlan = mlx4_en_set_vf_vlan,
2629 .ndo_set_vf_rate = mlx4_en_set_vf_rate,
2630 .ndo_set_vf_spoofchk = mlx4_en_set_vf_spoofchk,
2631 .ndo_set_vf_link_state = mlx4_en_set_vf_link_state,
2632 .ndo_get_vf_stats = mlx4_en_get_vf_stats,
2633 .ndo_get_vf_config = mlx4_en_get_vf_config,
2634 #ifdef CONFIG_NET_POLL_CONTROLLER
2635 .ndo_poll_controller = mlx4_en_netpoll,
2637 .ndo_set_features = mlx4_en_set_features,
2638 .ndo_fix_features = mlx4_en_fix_features,
2639 .ndo_setup_tc = __mlx4_en_setup_tc,
2640 #ifdef CONFIG_RFS_ACCEL
2641 .ndo_rx_flow_steer = mlx4_en_filter_rfs,
2643 .ndo_get_phys_port_id = mlx4_en_get_phys_port_id,
2644 #ifdef CONFIG_MLX4_EN_VXLAN
2645 .ndo_add_vxlan_port = mlx4_en_add_vxlan_port,
2646 .ndo_del_vxlan_port = mlx4_en_del_vxlan_port,
2647 .ndo_features_check = mlx4_en_features_check,
2649 .ndo_set_tx_maxrate = mlx4_en_set_tx_maxrate,
2652 struct mlx4_en_bond {
2653 struct work_struct work;
2654 struct mlx4_en_priv *priv;
2656 struct mlx4_port_map port_map;
2659 static void mlx4_en_bond_work(struct work_struct *work)
2661 struct mlx4_en_bond *bond = container_of(work,
2662 struct mlx4_en_bond,
2665 struct mlx4_dev *dev = bond->priv->mdev->dev;
2667 if (bond->is_bonded) {
2668 if (!mlx4_is_bonded(dev)) {
2669 err = mlx4_bond(dev);
2671 en_err(bond->priv, "Fail to bond device\n");
2674 err = mlx4_port_map_set(dev, &bond->port_map);
2676 en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2677 bond->port_map.port1,
2678 bond->port_map.port2,
2681 } else if (mlx4_is_bonded(dev)) {
2682 err = mlx4_unbond(dev);
2684 en_err(bond->priv, "Fail to unbond device\n");
2686 dev_put(bond->priv->dev);
2690 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2691 u8 v2p_p1, u8 v2p_p2)
2693 struct mlx4_en_bond *bond = NULL;
2695 bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2699 INIT_WORK(&bond->work, mlx4_en_bond_work);
2701 bond->is_bonded = is_bonded;
2702 bond->port_map.port1 = v2p_p1;
2703 bond->port_map.port2 = v2p_p2;
2704 dev_hold(priv->dev);
2705 queue_work(priv->mdev->workqueue, &bond->work);
2709 int mlx4_en_netdev_event(struct notifier_block *this,
2710 unsigned long event, void *ptr)
2712 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2714 struct mlx4_en_dev *mdev;
2715 struct mlx4_dev *dev;
2716 int i, num_eth_ports = 0;
2717 bool do_bond = true;
2718 struct mlx4_en_priv *priv;
2722 if (!net_eq(dev_net(ndev), &init_net))
2725 mdev = container_of(this, struct mlx4_en_dev, nb);
2728 /* Go into this mode only when two network devices set on two ports
2729 * of the same mlx4 device are slaves of the same bonding master
2731 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2733 if (!port && (mdev->pndev[i] == ndev))
2735 mdev->upper[i] = mdev->pndev[i] ?
2736 netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2737 /* condition not met: network device is a slave */
2738 if (!mdev->upper[i])
2740 if (num_eth_ports < 2)
2742 /* condition not met: same master */
2743 if (mdev->upper[i] != mdev->upper[i-1])
2746 /* condition not met: 2 salves */
2747 do_bond = (num_eth_ports == 2) ? do_bond : false;
2749 /* handle only events that come with enough info */
2750 if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
2753 priv = netdev_priv(ndev);
2755 struct netdev_notifier_bonding_info *notifier_info = ptr;
2756 struct netdev_bonding_info *bonding_info =
2757 ¬ifier_info->bonding_info;
2759 /* required mode 1, 2 or 4 */
2760 if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
2761 (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
2762 (bonding_info->master.bond_mode != BOND_MODE_8023AD))
2765 /* require exactly 2 slaves */
2766 if (bonding_info->master.num_slaves != 2)
2771 if (bonding_info->master.bond_mode ==
2772 BOND_MODE_ACTIVEBACKUP) {
2773 /* in active-backup mode virtual ports are
2774 * mapped to the physical port of the active
2776 if (bonding_info->slave.state ==
2777 BOND_STATE_BACKUP) {
2785 } else { /* BOND_STATE_ACTIVE */
2794 } else { /* Active-Active */
2795 /* in active-active mode a virtual port is
2796 * mapped to the native physical port if and only
2797 * if the physical port is up */
2798 __s8 link = bonding_info->slave.link;
2804 if ((link == BOND_LINK_UP) ||
2805 (link == BOND_LINK_FAIL)) {
2810 } else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
2820 mlx4_en_queue_bond_work(priv, do_bond,
2821 v2p_port1, v2p_port2);
2826 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
2827 struct mlx4_en_stats_bitmap *stats_bitmap,
2828 u8 rx_ppp, u8 rx_pause,
2829 u8 tx_ppp, u8 tx_pause)
2831 int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
2833 if (!mlx4_is_slave(dev) &&
2834 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
2835 mutex_lock(&stats_bitmap->mutex);
2836 bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
2839 bitmap_set(stats_bitmap->bitmap, last_i,
2840 NUM_FLOW_PRIORITY_STATS_RX);
2841 last_i += NUM_FLOW_PRIORITY_STATS_RX;
2843 if (rx_pause && !(rx_ppp))
2844 bitmap_set(stats_bitmap->bitmap, last_i,
2846 last_i += NUM_FLOW_STATS_RX;
2849 bitmap_set(stats_bitmap->bitmap, last_i,
2850 NUM_FLOW_PRIORITY_STATS_TX);
2851 last_i += NUM_FLOW_PRIORITY_STATS_TX;
2853 if (tx_pause && !(tx_ppp))
2854 bitmap_set(stats_bitmap->bitmap, last_i,
2856 last_i += NUM_FLOW_STATS_TX;
2858 mutex_unlock(&stats_bitmap->mutex);
2862 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
2863 struct mlx4_en_stats_bitmap *stats_bitmap,
2864 u8 rx_ppp, u8 rx_pause,
2865 u8 tx_ppp, u8 tx_pause)
2869 mutex_init(&stats_bitmap->mutex);
2870 bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
2872 if (mlx4_is_slave(dev)) {
2873 bitmap_set(stats_bitmap->bitmap, last_i +
2874 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
2875 bitmap_set(stats_bitmap->bitmap, last_i +
2876 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
2877 bitmap_set(stats_bitmap->bitmap, last_i +
2878 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
2879 bitmap_set(stats_bitmap->bitmap, last_i +
2880 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
2881 bitmap_set(stats_bitmap->bitmap, last_i +
2882 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
2883 bitmap_set(stats_bitmap->bitmap, last_i +
2884 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
2886 bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
2888 last_i += NUM_MAIN_STATS;
2890 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
2891 last_i += NUM_PORT_STATS;
2893 if (mlx4_is_master(dev))
2894 bitmap_set(stats_bitmap->bitmap, last_i,
2896 last_i += NUM_PF_STATS;
2898 mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
2901 last_i += NUM_FLOW_STATS;
2903 if (!mlx4_is_slave(dev))
2904 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
2907 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2908 struct mlx4_en_port_profile *prof)
2910 struct net_device *dev;
2911 struct mlx4_en_priv *priv;
2915 dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2916 MAX_TX_RINGS, MAX_RX_RINGS);
2920 netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2921 netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2923 SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
2924 dev->dev_port = port - 1;
2927 * Initialize driver private data
2930 priv = netdev_priv(dev);
2931 memset(priv, 0, sizeof(struct mlx4_en_priv));
2932 priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
2933 spin_lock_init(&priv->stats_lock);
2934 INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2935 INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2936 INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2937 INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2938 INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2939 #ifdef CONFIG_MLX4_EN_VXLAN
2940 INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
2941 INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
2943 #ifdef CONFIG_RFS_ACCEL
2944 INIT_LIST_HEAD(&priv->filters);
2945 spin_lock_init(&priv->filters_lock);
2950 priv->ddev = &mdev->pdev->dev;
2953 priv->port_up = false;
2954 priv->flags = prof->flags;
2955 priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
2956 priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2957 MLX4_WQE_CTRL_SOLICITED);
2958 priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2959 priv->tx_ring_num = prof->tx_ring_num;
2960 priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
2961 netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
2963 priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2965 if (!priv->tx_ring) {
2969 priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2975 priv->rx_ring_num = prof->rx_ring_num;
2976 priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2977 priv->cqe_size = mdev->dev->caps.cqe_size;
2978 priv->mac_index = -1;
2979 priv->msg_enable = MLX4_EN_MSG_LEVEL;
2980 #ifdef CONFIG_MLX4_EN_DCB
2981 if (!mlx4_is_slave(priv->mdev->dev)) {
2982 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
2983 dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2985 en_info(priv, "enabling only PFC DCB ops\n");
2986 dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2991 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2992 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2994 /* Query for default mac and max mtu */
2995 priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2997 if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
2998 MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
2999 priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3001 /* Set default MAC */
3002 dev->addr_len = ETH_ALEN;
3003 mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3004 if (!is_valid_ether_addr(dev->dev_addr)) {
3005 en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
3006 priv->port, dev->dev_addr);
3009 } else if (mlx4_is_slave(priv->mdev->dev) &&
3010 (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3011 /* Random MAC was assigned in mlx4_slave_cap
3012 * in mlx4_core module
3014 dev->addr_assign_type |= NET_ADDR_RANDOM;
3015 en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3018 memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3020 priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3021 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3022 err = mlx4_en_alloc_resources(priv);
3026 /* Initialize time stamping config */
3027 priv->hwtstamp_config.flags = 0;
3028 priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3029 priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3031 /* Allocate page for receive rings */
3032 err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3035 en_err(priv, "Failed to allocate page for rx qps\n");
3038 priv->allocated = 1;
3041 * Initialize netdev entry points
3043 if (mlx4_is_master(priv->mdev->dev))
3044 dev->netdev_ops = &mlx4_netdev_ops_master;
3046 dev->netdev_ops = &mlx4_netdev_ops;
3047 dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3048 netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
3049 netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3051 dev->ethtool_ops = &mlx4_en_ethtool_ops;
3054 * Set driver features
3056 dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3057 if (mdev->LSO_support)
3058 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3060 dev->vlan_features = dev->hw_features;
3062 dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3063 dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3064 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3065 NETIF_F_HW_VLAN_CTAG_FILTER;
3066 dev->hw_features |= NETIF_F_LOOPBACK |
3067 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3069 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3070 dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3071 NETIF_F_HW_VLAN_STAG_FILTER;
3072 dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3075 if (mlx4_is_slave(mdev->dev)) {
3078 err = get_phv_bit(mdev->dev, port, &phv);
3080 dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3081 priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3084 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3085 !(mdev->dev->caps.flags2 &
3086 MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3087 dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3090 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3091 dev->hw_features |= NETIF_F_RXFCS;
3093 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3094 dev->hw_features |= NETIF_F_RXALL;
3096 if (mdev->dev->caps.steering_mode ==
3097 MLX4_STEERING_MODE_DEVICE_MANAGED &&
3098 mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3099 dev->hw_features |= NETIF_F_NTUPLE;
3101 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3102 dev->priv_flags |= IFF_UNICAST_FLT;
3104 /* Setting a default hash function value */
3105 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3106 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3107 } else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3108 priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3111 "No RSS hash capabilities exposed, using Toeplitz\n");
3112 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3115 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3116 dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3117 NETIF_F_GSO_UDP_TUNNEL_CSUM |
3118 NETIF_F_GSO_PARTIAL;
3119 dev->features |= NETIF_F_GSO_UDP_TUNNEL |
3120 NETIF_F_GSO_UDP_TUNNEL_CSUM |
3121 NETIF_F_GSO_PARTIAL;
3122 dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3125 mdev->pndev[port] = dev;
3126 mdev->upper[port] = NULL;
3128 netif_carrier_off(dev);
3129 mlx4_en_set_default_moderation(priv);
3131 en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
3132 en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3134 mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3136 /* Configure port */
3137 mlx4_en_calc_rx_buf(dev);
3138 err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3139 priv->rx_skb_size + ETH_FCS_LEN,
3140 prof->tx_pause, prof->tx_ppp,
3141 prof->rx_pause, prof->rx_ppp);
3143 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3148 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3149 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3151 en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3158 en_warn(priv, "Initializing port\n");
3159 err = mlx4_INIT_PORT(mdev->dev, priv->port);
3161 en_err(priv, "Failed Initializing port\n");
3164 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3166 /* Initialize time stamp mechanism */
3167 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3168 mlx4_en_init_timestamp(mdev);
3170 queue_delayed_work(mdev->workqueue, &priv->service_task,
3171 SERVICE_TASK_DELAY);
3173 mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3174 mdev->profile.prof[priv->port].rx_ppp,
3175 mdev->profile.prof[priv->port].rx_pause,
3176 mdev->profile.prof[priv->port].tx_ppp,
3177 mdev->profile.prof[priv->port].tx_pause);
3179 err = register_netdev(dev);
3181 en_err(priv, "Netdev registration failed for port %d\n", port);
3185 priv->registered = 1;
3186 devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3192 mlx4_en_destroy_netdev(dev);
3196 int mlx4_en_reset_config(struct net_device *dev,
3197 struct hwtstamp_config ts_config,
3198 netdev_features_t features)
3200 struct mlx4_en_priv *priv = netdev_priv(dev);
3201 struct mlx4_en_dev *mdev = priv->mdev;
3202 struct mlx4_en_port_profile new_prof;
3203 struct mlx4_en_priv *tmp;
3207 if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3208 priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3209 !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3210 !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3211 return 0; /* Nothing to change */
3213 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3214 (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3215 (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3216 en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3220 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3224 mutex_lock(&mdev->state_lock);
3226 memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3227 memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3229 err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof);
3233 if (priv->port_up) {
3235 mlx4_en_stop_port(dev, 1);
3238 en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
3239 ts_config.rx_filter,
3240 !!(features & NETIF_F_HW_VLAN_CTAG_RX));
3242 mlx4_en_safe_replace_resources(priv, tmp);
3244 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3245 if (features & NETIF_F_HW_VLAN_CTAG_RX)
3246 dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3248 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3249 } else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3250 /* RX time-stamping is OFF, update the RX vlan offload
3251 * to the latest wanted state
3253 if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3254 dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3256 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3259 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3260 if (features & NETIF_F_RXFCS)
3261 dev->features |= NETIF_F_RXFCS;
3263 dev->features &= ~NETIF_F_RXFCS;
3266 /* RX vlan offload and RX time-stamping can't co-exist !
3267 * Regardless of the caller's choice,
3268 * Turn Off RX vlan offload in case of time-stamping is ON
3270 if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3271 if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3272 en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3273 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3277 err = mlx4_en_start_port(dev);
3279 en_err(priv, "Failed starting port\n");
3283 mutex_unlock(&mdev->state_lock);
3286 netdev_features_change(dev);