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/bpf.h>
35 #include <linux/etherdevice.h>
36 #include <linux/tcp.h>
37 #include <linux/if_vlan.h>
38 #include <linux/delay.h>
39 #include <linux/slab.h>
40 #include <linux/hash.h>
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 #define MLX4_EN_MAX_XDP_MTU ((int)(PAGE_SIZE - ETH_HLEN - (2 * VLAN_HLEN) - \
56 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
58 struct mlx4_en_priv *priv = netdev_priv(dev);
60 unsigned int offset = 0;
62 if (up && up != MLX4_EN_NUM_UP_HIGH)
65 netdev_set_num_tc(dev, up);
66 netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
67 /* Partition Tx queues evenly amongst UP's */
68 for (i = 0; i < up; i++) {
69 netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
70 offset += priv->num_tx_rings_p_up;
73 #ifdef CONFIG_MLX4_EN_DCB
74 if (!mlx4_is_slave(priv->mdev->dev)) {
77 priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
79 priv->flags &= ~MLX4_EN_FLAG_DCB_ENABLED;
80 priv->cee_config.pfc_state = false;
83 #endif /* CONFIG_MLX4_EN_DCB */
88 int mlx4_en_alloc_tx_queue_per_tc(struct net_device *dev, u8 tc)
90 struct mlx4_en_priv *priv = netdev_priv(dev);
91 struct mlx4_en_dev *mdev = priv->mdev;
92 struct mlx4_en_port_profile new_prof;
93 struct mlx4_en_priv *tmp;
97 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
101 mutex_lock(&mdev->state_lock);
102 memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
103 new_prof.num_up = (tc == 0) ? MLX4_EN_NUM_UP_LOW :
105 new_prof.tx_ring_num[TX] = new_prof.num_tx_rings_p_up *
107 err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
113 mlx4_en_stop_port(dev, 1);
116 mlx4_en_safe_replace_resources(priv, tmp);
118 err = mlx4_en_start_port(dev);
120 en_err(priv, "Failed starting port for setup TC\n");
125 err = mlx4_en_setup_tc(dev, tc);
127 mutex_unlock(&mdev->state_lock);
132 static int __mlx4_en_setup_tc(struct net_device *dev, enum tc_setup_type type,
135 struct tc_mqprio_qopt *mqprio = type_data;
137 if (type != TC_SETUP_QDISC_MQPRIO)
140 if (mqprio->num_tc && mqprio->num_tc != MLX4_EN_NUM_UP_HIGH)
143 mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
145 return mlx4_en_alloc_tx_queue_per_tc(dev, mqprio->num_tc);
148 #ifdef CONFIG_RFS_ACCEL
150 struct mlx4_en_filter {
151 struct list_head next;
152 struct work_struct work;
161 struct mlx4_en_priv *priv;
162 u32 flow_id; /* RFS infrastructure id */
163 int id; /* mlx4_en driver id */
164 u64 reg_id; /* Flow steering API id */
165 u8 activated; /* Used to prevent expiry before filter
168 struct hlist_node filter_chain;
171 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
173 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
177 return MLX4_NET_TRANS_RULE_ID_UDP;
179 return MLX4_NET_TRANS_RULE_ID_TCP;
181 return MLX4_NET_TRANS_RULE_NUM;
185 /* Must not acquire state_lock, as its corresponding work_sync
188 static void mlx4_en_filter_work(struct work_struct *work)
190 struct mlx4_en_filter *filter = container_of(work,
191 struct mlx4_en_filter,
193 struct mlx4_en_priv *priv = filter->priv;
194 struct mlx4_spec_list spec_tcp_udp = {
195 .id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
198 .dst_port = filter->dst_port,
199 .dst_port_msk = (__force __be16)-1,
200 .src_port = filter->src_port,
201 .src_port_msk = (__force __be16)-1,
205 struct mlx4_spec_list spec_ip = {
206 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
209 .dst_ip = filter->dst_ip,
210 .dst_ip_msk = (__force __be32)-1,
211 .src_ip = filter->src_ip,
212 .src_ip_msk = (__force __be32)-1,
216 struct mlx4_spec_list spec_eth = {
217 .id = MLX4_NET_TRANS_RULE_ID_ETH,
219 struct mlx4_net_trans_rule rule = {
220 .list = LIST_HEAD_INIT(rule.list),
221 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
224 .promisc_mode = MLX4_FS_REGULAR,
226 .priority = MLX4_DOMAIN_RFS,
229 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
231 if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
232 en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
236 list_add_tail(&spec_eth.list, &rule.list);
237 list_add_tail(&spec_ip.list, &rule.list);
238 list_add_tail(&spec_tcp_udp.list, &rule.list);
240 rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
241 memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
242 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
244 filter->activated = 0;
246 if (filter->reg_id) {
247 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
248 if (rc && rc != -ENOENT)
249 en_err(priv, "Error detaching flow. rc = %d\n", rc);
252 rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
254 en_err(priv, "Error attaching flow. err = %d\n", rc);
257 mlx4_en_filter_rfs_expire(priv);
259 filter->activated = 1;
262 static inline struct hlist_head *
263 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
264 __be16 src_port, __be16 dst_port)
269 l = (__force unsigned long)src_port |
270 ((__force unsigned long)dst_port << 2);
271 l ^= (__force unsigned long)(src_ip ^ dst_ip);
273 bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
275 return &priv->filter_hash[bucket_idx];
278 static struct mlx4_en_filter *
279 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
280 __be32 dst_ip, u8 ip_proto, __be16 src_port,
281 __be16 dst_port, u32 flow_id)
283 struct mlx4_en_filter *filter = NULL;
285 filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
290 filter->rxq_index = rxq_index;
291 INIT_WORK(&filter->work, mlx4_en_filter_work);
293 filter->src_ip = src_ip;
294 filter->dst_ip = dst_ip;
295 filter->ip_proto = ip_proto;
296 filter->src_port = src_port;
297 filter->dst_port = dst_port;
299 filter->flow_id = flow_id;
301 filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
303 list_add_tail(&filter->next, &priv->filters);
304 hlist_add_head(&filter->filter_chain,
305 filter_hash_bucket(priv, src_ip, dst_ip, src_port,
311 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
313 struct mlx4_en_priv *priv = filter->priv;
316 list_del(&filter->next);
318 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
319 if (rc && rc != -ENOENT)
320 en_err(priv, "Error detaching flow. rc = %d\n", rc);
325 static inline struct mlx4_en_filter *
326 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
327 u8 ip_proto, __be16 src_port, __be16 dst_port)
329 struct mlx4_en_filter *filter;
330 struct mlx4_en_filter *ret = NULL;
332 hlist_for_each_entry(filter,
333 filter_hash_bucket(priv, src_ip, dst_ip,
336 if (filter->src_ip == src_ip &&
337 filter->dst_ip == dst_ip &&
338 filter->ip_proto == ip_proto &&
339 filter->src_port == src_port &&
340 filter->dst_port == dst_port) {
350 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
351 u16 rxq_index, u32 flow_id)
353 struct mlx4_en_priv *priv = netdev_priv(net_dev);
354 struct mlx4_en_filter *filter;
355 const struct iphdr *ip;
362 int nhoff = skb_network_offset(skb);
365 if (skb->protocol != htons(ETH_P_IP))
366 return -EPROTONOSUPPORT;
368 ip = (const struct iphdr *)(skb->data + nhoff);
369 if (ip_is_fragment(ip))
370 return -EPROTONOSUPPORT;
372 if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
373 return -EPROTONOSUPPORT;
374 ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
376 ip_proto = ip->protocol;
382 spin_lock_bh(&priv->filters_lock);
383 filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
386 if (filter->rxq_index == rxq_index)
389 filter->rxq_index = rxq_index;
391 filter = mlx4_en_filter_alloc(priv, rxq_index,
392 src_ip, dst_ip, ip_proto,
393 src_port, dst_port, flow_id);
400 queue_work(priv->mdev->workqueue, &filter->work);
405 spin_unlock_bh(&priv->filters_lock);
410 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
412 struct mlx4_en_filter *filter, *tmp;
415 spin_lock_bh(&priv->filters_lock);
416 list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
417 list_move(&filter->next, &del_list);
418 hlist_del(&filter->filter_chain);
420 spin_unlock_bh(&priv->filters_lock);
422 list_for_each_entry_safe(filter, tmp, &del_list, next) {
423 cancel_work_sync(&filter->work);
424 mlx4_en_filter_free(filter);
428 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
430 struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
434 spin_lock_bh(&priv->filters_lock);
435 list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
436 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
439 if (filter->activated &&
440 !work_pending(&filter->work) &&
441 rps_may_expire_flow(priv->dev,
442 filter->rxq_index, filter->flow_id,
444 list_move(&filter->next, &del_list);
445 hlist_del(&filter->filter_chain);
447 last_filter = filter;
452 if (last_filter && (&last_filter->next != priv->filters.next))
453 list_move(&priv->filters, &last_filter->next);
455 spin_unlock_bh(&priv->filters_lock);
457 list_for_each_entry_safe(filter, tmp, &del_list, next)
458 mlx4_en_filter_free(filter);
462 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
463 __be16 proto, u16 vid)
465 struct mlx4_en_priv *priv = netdev_priv(dev);
466 struct mlx4_en_dev *mdev = priv->mdev;
470 en_dbg(HW, priv, "adding VLAN:%d\n", vid);
472 set_bit(vid, priv->active_vlans);
474 /* Add VID to port VLAN filter */
475 mutex_lock(&mdev->state_lock);
476 if (mdev->device_up && priv->port_up) {
477 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
479 en_err(priv, "Failed configuring VLAN filter\n");
483 err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
485 en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
488 mutex_unlock(&mdev->state_lock);
492 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
493 __be16 proto, u16 vid)
495 struct mlx4_en_priv *priv = netdev_priv(dev);
496 struct mlx4_en_dev *mdev = priv->mdev;
499 en_dbg(HW, priv, "Killing VID:%d\n", vid);
501 clear_bit(vid, priv->active_vlans);
503 /* Remove VID from port VLAN filter */
504 mutex_lock(&mdev->state_lock);
505 mlx4_unregister_vlan(mdev->dev, priv->port, vid);
507 if (mdev->device_up && priv->port_up) {
508 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
510 en_err(priv, "Failed configuring VLAN filter\n");
512 mutex_unlock(&mdev->state_lock);
517 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
520 for (i = ETH_ALEN - 1; i >= 0; --i) {
521 dst_mac[i] = src_mac & 0xff;
524 memset(&dst_mac[ETH_ALEN], 0, 2);
528 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
529 int qpn, u64 *reg_id)
533 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
534 priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
535 return 0; /* do nothing */
537 err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
538 MLX4_DOMAIN_NIC, reg_id);
540 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
543 en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
548 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
549 unsigned char *mac, int *qpn, u64 *reg_id)
551 struct mlx4_en_dev *mdev = priv->mdev;
552 struct mlx4_dev *dev = mdev->dev;
555 switch (dev->caps.steering_mode) {
556 case MLX4_STEERING_MODE_B0: {
561 memcpy(&gid[10], mac, ETH_ALEN);
564 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
567 case MLX4_STEERING_MODE_DEVICE_MANAGED: {
568 struct mlx4_spec_list spec_eth = { {NULL} };
569 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
571 struct mlx4_net_trans_rule rule = {
572 .queue_mode = MLX4_NET_TRANS_Q_FIFO,
575 .promisc_mode = MLX4_FS_REGULAR,
576 .priority = MLX4_DOMAIN_NIC,
579 rule.port = priv->port;
581 INIT_LIST_HEAD(&rule.list);
583 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
584 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
585 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
586 list_add_tail(&spec_eth.list, &rule.list);
588 err = mlx4_flow_attach(dev, &rule, reg_id);
595 en_warn(priv, "Failed Attaching Unicast\n");
600 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
601 unsigned char *mac, int qpn, u64 reg_id)
603 struct mlx4_en_dev *mdev = priv->mdev;
604 struct mlx4_dev *dev = mdev->dev;
606 switch (dev->caps.steering_mode) {
607 case MLX4_STEERING_MODE_B0: {
612 memcpy(&gid[10], mac, ETH_ALEN);
615 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
618 case MLX4_STEERING_MODE_DEVICE_MANAGED: {
619 mlx4_flow_detach(dev, reg_id);
623 en_err(priv, "Invalid steering mode.\n");
627 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
629 struct mlx4_en_dev *mdev = priv->mdev;
630 struct mlx4_dev *dev = mdev->dev;
633 int *qpn = &priv->base_qpn;
634 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
636 en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
637 priv->dev->dev_addr);
638 index = mlx4_register_mac(dev, priv->port, mac);
641 en_err(priv, "Failed adding MAC: %pM\n",
642 priv->dev->dev_addr);
646 en_info(priv, "Steering Mode %d\n", dev->caps.steering_mode);
648 if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
649 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
650 *qpn = base_qpn + index;
654 err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP,
655 MLX4_RES_USAGE_DRIVER);
656 en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
658 en_err(priv, "Failed to reserve qp for mac registration\n");
659 mlx4_unregister_mac(dev, priv->port, mac);
666 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
668 struct mlx4_en_dev *mdev = priv->mdev;
669 struct mlx4_dev *dev = mdev->dev;
670 int qpn = priv->base_qpn;
672 if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
673 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
674 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
675 priv->dev->dev_addr);
676 mlx4_unregister_mac(dev, priv->port, mac);
678 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
680 mlx4_qp_release_range(dev, qpn, 1);
681 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
685 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
686 unsigned char *new_mac, unsigned char *prev_mac)
688 struct mlx4_en_dev *mdev = priv->mdev;
689 struct mlx4_dev *dev = mdev->dev;
691 u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
693 if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
694 struct hlist_head *bucket;
695 unsigned int mac_hash;
696 struct mlx4_mac_entry *entry;
697 struct hlist_node *tmp;
698 u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
700 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
701 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
702 if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
703 mlx4_en_uc_steer_release(priv, entry->mac,
705 mlx4_unregister_mac(dev, priv->port,
707 hlist_del_rcu(&entry->hlist);
709 memcpy(entry->mac, new_mac, ETH_ALEN);
711 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
712 hlist_add_head_rcu(&entry->hlist,
713 &priv->mac_hash[mac_hash]);
714 mlx4_register_mac(dev, priv->port, new_mac_u64);
715 err = mlx4_en_uc_steer_add(priv, new_mac,
720 if (priv->tunnel_reg_id) {
721 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
722 priv->tunnel_reg_id = 0;
724 err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
725 &priv->tunnel_reg_id);
732 return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
735 static void mlx4_en_update_user_mac(struct mlx4_en_priv *priv,
736 unsigned char new_mac[ETH_ALEN + 2])
738 struct mlx4_en_dev *mdev = priv->mdev;
741 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_USER_MAC_EN))
744 err = mlx4_SET_PORT_user_mac(mdev->dev, priv->port, new_mac);
746 en_err(priv, "Failed to pass user MAC(%pM) to Firmware for port %d, with error %d\n",
747 new_mac, priv->port, err);
750 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
751 unsigned char new_mac[ETH_ALEN + 2])
756 /* Remove old MAC and insert the new one */
757 err = mlx4_en_replace_mac(priv, priv->base_qpn,
758 new_mac, priv->current_mac);
760 en_err(priv, "Failed changing HW MAC address\n");
762 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
765 memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
770 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
772 struct mlx4_en_priv *priv = netdev_priv(dev);
773 struct mlx4_en_dev *mdev = priv->mdev;
774 struct sockaddr *saddr = addr;
775 unsigned char new_mac[ETH_ALEN + 2];
778 if (!is_valid_ether_addr(saddr->sa_data))
779 return -EADDRNOTAVAIL;
781 mutex_lock(&mdev->state_lock);
782 memcpy(new_mac, saddr->sa_data, ETH_ALEN);
783 err = mlx4_en_do_set_mac(priv, new_mac);
787 memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
788 mlx4_en_update_user_mac(priv, new_mac);
790 mutex_unlock(&mdev->state_lock);
795 static void mlx4_en_clear_list(struct net_device *dev)
797 struct mlx4_en_priv *priv = netdev_priv(dev);
798 struct mlx4_en_mc_list *tmp, *mc_to_del;
800 list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
801 list_del(&mc_to_del->list);
806 static void mlx4_en_cache_mclist(struct net_device *dev)
808 struct mlx4_en_priv *priv = netdev_priv(dev);
809 struct netdev_hw_addr *ha;
810 struct mlx4_en_mc_list *tmp;
812 mlx4_en_clear_list(dev);
813 netdev_for_each_mc_addr(ha, dev) {
814 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
816 mlx4_en_clear_list(dev);
819 memcpy(tmp->addr, ha->addr, ETH_ALEN);
820 list_add_tail(&tmp->list, &priv->mc_list);
824 static void update_mclist_flags(struct mlx4_en_priv *priv,
825 struct list_head *dst,
826 struct list_head *src)
828 struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
831 /* Find all the entries that should be removed from dst,
832 * These are the entries that are not found in src
834 list_for_each_entry(dst_tmp, dst, list) {
836 list_for_each_entry(src_tmp, src, list) {
837 if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
843 dst_tmp->action = MCLIST_REM;
846 /* Add entries that exist in src but not in dst
847 * mark them as need to add
849 list_for_each_entry(src_tmp, src, list) {
851 list_for_each_entry(dst_tmp, dst, list) {
852 if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
853 dst_tmp->action = MCLIST_NONE;
859 new_mc = kmemdup(src_tmp,
860 sizeof(struct mlx4_en_mc_list),
865 new_mc->action = MCLIST_ADD;
866 list_add_tail(&new_mc->list, dst);
871 static void mlx4_en_set_rx_mode(struct net_device *dev)
873 struct mlx4_en_priv *priv = netdev_priv(dev);
878 queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
881 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
882 struct mlx4_en_dev *mdev)
886 if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
887 if (netif_msg_rx_status(priv))
888 en_warn(priv, "Entering promiscuous mode\n");
889 priv->flags |= MLX4_EN_FLAG_PROMISC;
891 /* Enable promiscouos mode */
892 switch (mdev->dev->caps.steering_mode) {
893 case MLX4_STEERING_MODE_DEVICE_MANAGED:
894 err = mlx4_flow_steer_promisc_add(mdev->dev,
897 MLX4_FS_ALL_DEFAULT);
899 en_err(priv, "Failed enabling promiscuous mode\n");
900 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
903 case MLX4_STEERING_MODE_B0:
904 err = mlx4_unicast_promisc_add(mdev->dev,
908 en_err(priv, "Failed enabling unicast promiscuous mode\n");
910 /* Add the default qp number as multicast
913 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
914 err = mlx4_multicast_promisc_add(mdev->dev,
918 en_err(priv, "Failed enabling multicast promiscuous mode\n");
919 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
923 case MLX4_STEERING_MODE_A0:
924 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
929 en_err(priv, "Failed enabling promiscuous mode\n");
933 /* Disable port multicast filter (unconditionally) */
934 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
935 0, MLX4_MCAST_DISABLE);
937 en_err(priv, "Failed disabling multicast filter\n");
941 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
942 struct mlx4_en_dev *mdev)
946 if (netif_msg_rx_status(priv))
947 en_warn(priv, "Leaving promiscuous mode\n");
948 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
950 /* Disable promiscouos mode */
951 switch (mdev->dev->caps.steering_mode) {
952 case MLX4_STEERING_MODE_DEVICE_MANAGED:
953 err = mlx4_flow_steer_promisc_remove(mdev->dev,
955 MLX4_FS_ALL_DEFAULT);
957 en_err(priv, "Failed disabling promiscuous mode\n");
958 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
961 case MLX4_STEERING_MODE_B0:
962 err = mlx4_unicast_promisc_remove(mdev->dev,
966 en_err(priv, "Failed disabling unicast promiscuous mode\n");
967 /* Disable Multicast promisc */
968 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
969 err = mlx4_multicast_promisc_remove(mdev->dev,
973 en_err(priv, "Failed disabling multicast promiscuous mode\n");
974 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
978 case MLX4_STEERING_MODE_A0:
979 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
983 en_err(priv, "Failed disabling promiscuous mode\n");
988 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
989 struct net_device *dev,
990 struct mlx4_en_dev *mdev)
992 struct mlx4_en_mc_list *mclist, *tmp;
994 u8 mc_list[16] = {0};
997 /* Enable/disable the multicast filter according to IFF_ALLMULTI */
998 if (dev->flags & IFF_ALLMULTI) {
999 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1000 0, MLX4_MCAST_DISABLE);
1002 en_err(priv, "Failed disabling multicast filter\n");
1004 /* Add the default qp number as multicast promisc */
1005 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1006 switch (mdev->dev->caps.steering_mode) {
1007 case MLX4_STEERING_MODE_DEVICE_MANAGED:
1008 err = mlx4_flow_steer_promisc_add(mdev->dev,
1011 MLX4_FS_MC_DEFAULT);
1014 case MLX4_STEERING_MODE_B0:
1015 err = mlx4_multicast_promisc_add(mdev->dev,
1020 case MLX4_STEERING_MODE_A0:
1024 en_err(priv, "Failed entering multicast promisc mode\n");
1025 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1028 /* Disable Multicast promisc */
1029 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1030 switch (mdev->dev->caps.steering_mode) {
1031 case MLX4_STEERING_MODE_DEVICE_MANAGED:
1032 err = mlx4_flow_steer_promisc_remove(mdev->dev,
1034 MLX4_FS_MC_DEFAULT);
1037 case MLX4_STEERING_MODE_B0:
1038 err = mlx4_multicast_promisc_remove(mdev->dev,
1043 case MLX4_STEERING_MODE_A0:
1047 en_err(priv, "Failed disabling multicast promiscuous mode\n");
1048 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1051 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1052 0, MLX4_MCAST_DISABLE);
1054 en_err(priv, "Failed disabling multicast filter\n");
1056 /* Flush mcast filter and init it with broadcast address */
1057 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1058 1, MLX4_MCAST_CONFIG);
1060 /* Update multicast list - we cache all addresses so they won't
1061 * change while HW is updated holding the command semaphor */
1062 netif_addr_lock_bh(dev);
1063 mlx4_en_cache_mclist(dev);
1064 netif_addr_unlock_bh(dev);
1065 list_for_each_entry(mclist, &priv->mc_list, list) {
1066 mcast_addr = mlx4_mac_to_u64(mclist->addr);
1067 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1068 mcast_addr, 0, MLX4_MCAST_CONFIG);
1070 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1071 0, MLX4_MCAST_ENABLE);
1073 en_err(priv, "Failed enabling multicast filter\n");
1075 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1076 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1077 if (mclist->action == MCLIST_REM) {
1078 /* detach this address and delete from list */
1079 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1080 mc_list[5] = priv->port;
1081 err = mlx4_multicast_detach(mdev->dev,
1082 priv->rss_map.indir_qp,
1087 en_err(priv, "Fail to detach multicast address\n");
1089 if (mclist->tunnel_reg_id) {
1090 err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1092 en_err(priv, "Failed to detach multicast address\n");
1095 /* remove from list */
1096 list_del(&mclist->list);
1098 } else if (mclist->action == MCLIST_ADD) {
1099 /* attach the address */
1100 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1101 /* needed for B0 steering support */
1102 mc_list[5] = priv->port;
1103 err = mlx4_multicast_attach(mdev->dev,
1104 priv->rss_map.indir_qp,
1110 en_err(priv, "Fail to attach multicast address\n");
1112 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1113 &mclist->tunnel_reg_id);
1115 en_err(priv, "Failed to attach multicast address\n");
1121 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1122 struct net_device *dev,
1123 struct mlx4_en_dev *mdev)
1125 struct netdev_hw_addr *ha;
1126 struct mlx4_mac_entry *entry;
1127 struct hlist_node *tmp;
1131 struct hlist_head *bucket;
1136 /* Note that we do not need to protect our mac_hash traversal with rcu,
1137 * since all modification code is protected by mdev->state_lock
1140 /* find what to remove */
1141 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1142 bucket = &priv->mac_hash[i];
1143 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1145 netdev_for_each_uc_addr(ha, dev) {
1146 if (ether_addr_equal_64bits(entry->mac,
1153 /* MAC address of the port is not in uc list */
1154 if (ether_addr_equal_64bits(entry->mac,
1159 mac = mlx4_mac_to_u64(entry->mac);
1160 mlx4_en_uc_steer_release(priv, entry->mac,
1163 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1165 hlist_del_rcu(&entry->hlist);
1166 kfree_rcu(entry, rcu);
1167 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1168 entry->mac, priv->port);
1174 /* if we didn't remove anything, there is no use in trying to add
1175 * again once we are in a forced promisc mode state
1177 if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1180 prev_flags = priv->flags;
1181 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1183 /* find what to add */
1184 netdev_for_each_uc_addr(ha, dev) {
1186 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1187 hlist_for_each_entry(entry, bucket, hlist) {
1188 if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1195 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1197 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1198 ha->addr, priv->port);
1199 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1202 mac = mlx4_mac_to_u64(ha->addr);
1203 memcpy(entry->mac, ha->addr, ETH_ALEN);
1204 err = mlx4_register_mac(mdev->dev, priv->port, mac);
1206 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1207 ha->addr, priv->port, err);
1209 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1212 err = mlx4_en_uc_steer_add(priv, ha->addr,
1216 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1217 ha->addr, priv->port, err);
1218 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1220 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1223 unsigned int mac_hash;
1224 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1225 ha->addr, priv->port);
1226 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1227 bucket = &priv->mac_hash[mac_hash];
1228 hlist_add_head_rcu(&entry->hlist, bucket);
1233 if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1234 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1236 } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1237 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1242 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1244 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1246 struct mlx4_en_dev *mdev = priv->mdev;
1247 struct net_device *dev = priv->dev;
1249 mutex_lock(&mdev->state_lock);
1250 if (!mdev->device_up) {
1251 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1254 if (!priv->port_up) {
1255 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1259 if (!netif_carrier_ok(dev)) {
1260 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1261 if (priv->port_state.link_state) {
1262 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1263 netif_carrier_on(dev);
1264 en_dbg(LINK, priv, "Link Up\n");
1269 if (dev->priv_flags & IFF_UNICAST_FLT)
1270 mlx4_en_do_uc_filter(priv, dev, mdev);
1272 /* Promsicuous mode: disable all filters */
1273 if ((dev->flags & IFF_PROMISC) ||
1274 (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1275 mlx4_en_set_promisc_mode(priv, mdev);
1279 /* Not in promiscuous mode */
1280 if (priv->flags & MLX4_EN_FLAG_PROMISC)
1281 mlx4_en_clear_promisc_mode(priv, mdev);
1283 mlx4_en_do_multicast(priv, dev, mdev);
1285 mutex_unlock(&mdev->state_lock);
1288 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1292 int *qpn = &priv->base_qpn;
1293 struct mlx4_mac_entry *entry;
1295 err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, ®_id);
1299 err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1300 &priv->tunnel_reg_id);
1304 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1310 memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1311 memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1312 entry->reg_id = reg_id;
1313 hlist_add_head_rcu(&entry->hlist,
1314 &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1319 if (priv->tunnel_reg_id)
1320 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1323 mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1327 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1331 int qpn = priv->base_qpn;
1332 struct hlist_head *bucket;
1333 struct hlist_node *tmp;
1334 struct mlx4_mac_entry *entry;
1336 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1337 bucket = &priv->mac_hash[i];
1338 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1339 mac = mlx4_mac_to_u64(entry->mac);
1340 en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1342 mlx4_en_uc_steer_release(priv, entry->mac,
1343 qpn, entry->reg_id);
1345 mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1346 hlist_del_rcu(&entry->hlist);
1347 kfree_rcu(entry, rcu);
1351 if (priv->tunnel_reg_id) {
1352 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1353 priv->tunnel_reg_id = 0;
1357 static void mlx4_en_tx_timeout(struct net_device *dev)
1359 struct mlx4_en_priv *priv = netdev_priv(dev);
1360 struct mlx4_en_dev *mdev = priv->mdev;
1363 if (netif_msg_timer(priv))
1364 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1366 for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1367 struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][i];
1369 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1371 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1372 i, tx_ring->qpn, tx_ring->sp_cqn,
1373 tx_ring->cons, tx_ring->prod);
1376 priv->port_stats.tx_timeout++;
1377 en_dbg(DRV, priv, "Scheduling watchdog\n");
1378 queue_work(mdev->workqueue, &priv->watchdog_task);
1383 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1385 struct mlx4_en_priv *priv = netdev_priv(dev);
1387 spin_lock_bh(&priv->stats_lock);
1388 mlx4_en_fold_software_stats(dev);
1389 netdev_stats_to_stats64(stats, &dev->stats);
1390 spin_unlock_bh(&priv->stats_lock);
1393 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1395 struct mlx4_en_cq *cq;
1398 /* If we haven't received a specific coalescing setting
1399 * (module param), we set the moderation parameters as follows:
1400 * - moder_cnt is set to the number of mtu sized packets to
1401 * satisfy our coalescing target.
1402 * - moder_time is set to a fixed value.
1404 priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1405 priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1406 priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1407 priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1408 en_dbg(INTR, priv, "Default coalescing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1409 priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1411 /* Setup cq moderation params */
1412 for (i = 0; i < priv->rx_ring_num; i++) {
1413 cq = priv->rx_cq[i];
1414 cq->moder_cnt = priv->rx_frames;
1415 cq->moder_time = priv->rx_usecs;
1416 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1417 priv->last_moder_packets[i] = 0;
1418 priv->last_moder_bytes[i] = 0;
1421 for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1422 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1423 cq = priv->tx_cq[t][i];
1424 cq->moder_cnt = priv->tx_frames;
1425 cq->moder_time = priv->tx_usecs;
1429 /* Reset auto-moderation params */
1430 priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1431 priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1432 priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1433 priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1434 priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1435 priv->adaptive_rx_coal = 1;
1436 priv->last_moder_jiffies = 0;
1437 priv->last_moder_tx_packets = 0;
1440 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1442 unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1443 u32 pkt_rate_high, pkt_rate_low;
1444 struct mlx4_en_cq *cq;
1445 unsigned long packets;
1447 unsigned long avg_pkt_size;
1448 unsigned long rx_packets;
1449 unsigned long rx_bytes;
1450 unsigned long rx_pkt_diff;
1454 if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1457 pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1458 pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1460 for (ring = 0; ring < priv->rx_ring_num; ring++) {
1461 rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1462 rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1464 rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1465 packets = rx_pkt_diff;
1466 rate = packets * HZ / period;
1467 avg_pkt_size = packets ? (rx_bytes -
1468 priv->last_moder_bytes[ring]) / packets : 0;
1470 /* Apply auto-moderation only when packet rate
1471 * exceeds a rate that it matters */
1472 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1473 avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1474 if (rate <= pkt_rate_low)
1475 moder_time = priv->rx_usecs_low;
1476 else if (rate >= pkt_rate_high)
1477 moder_time = priv->rx_usecs_high;
1479 moder_time = (rate - pkt_rate_low) *
1480 (priv->rx_usecs_high - priv->rx_usecs_low) /
1481 (pkt_rate_high - pkt_rate_low) +
1484 moder_time = priv->rx_usecs_low;
1487 cq = priv->rx_cq[ring];
1488 if (moder_time != priv->last_moder_time[ring] ||
1489 cq->moder_cnt != priv->rx_frames) {
1490 priv->last_moder_time[ring] = moder_time;
1491 cq->moder_time = moder_time;
1492 cq->moder_cnt = priv->rx_frames;
1493 err = mlx4_en_set_cq_moder(priv, cq);
1495 en_err(priv, "Failed modifying moderation for cq:%d\n",
1498 priv->last_moder_packets[ring] = rx_packets;
1499 priv->last_moder_bytes[ring] = rx_bytes;
1502 priv->last_moder_jiffies = jiffies;
1505 static void mlx4_en_do_get_stats(struct work_struct *work)
1507 struct delayed_work *delay = to_delayed_work(work);
1508 struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1510 struct mlx4_en_dev *mdev = priv->mdev;
1513 mutex_lock(&mdev->state_lock);
1514 if (mdev->device_up) {
1515 if (priv->port_up) {
1516 err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1518 en_dbg(HW, priv, "Could not update stats\n");
1520 mlx4_en_auto_moderation(priv);
1523 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1525 if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1526 mlx4_en_do_set_mac(priv, priv->current_mac);
1527 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1529 mutex_unlock(&mdev->state_lock);
1532 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1535 static void mlx4_en_service_task(struct work_struct *work)
1537 struct delayed_work *delay = to_delayed_work(work);
1538 struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1540 struct mlx4_en_dev *mdev = priv->mdev;
1542 mutex_lock(&mdev->state_lock);
1543 if (mdev->device_up) {
1544 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1545 mlx4_en_ptp_overflow_check(mdev);
1547 mlx4_en_recover_from_oom(priv);
1548 queue_delayed_work(mdev->workqueue, &priv->service_task,
1549 SERVICE_TASK_DELAY);
1551 mutex_unlock(&mdev->state_lock);
1554 static void mlx4_en_linkstate(struct work_struct *work)
1556 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1558 struct mlx4_en_dev *mdev = priv->mdev;
1559 int linkstate = priv->link_state;
1561 mutex_lock(&mdev->state_lock);
1562 /* If observable port state changed set carrier state and
1563 * report to system log */
1564 if (priv->last_link_state != linkstate) {
1565 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1566 en_info(priv, "Link Down\n");
1567 netif_carrier_off(priv->dev);
1569 en_info(priv, "Link Up\n");
1570 netif_carrier_on(priv->dev);
1573 priv->last_link_state = linkstate;
1574 mutex_unlock(&mdev->state_lock);
1577 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1579 struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1580 int numa_node = priv->mdev->dev->numa_node;
1582 if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1585 cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1586 ring->affinity_mask);
1590 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1592 free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1595 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1598 struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1599 int rr_index = tx_ring_idx;
1601 tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1602 tx_ring->recycle_ring = priv->rx_ring[rr_index];
1603 en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1604 TX_XDP, tx_ring_idx, rr_index);
1607 int mlx4_en_start_port(struct net_device *dev)
1609 struct mlx4_en_priv *priv = netdev_priv(dev);
1610 struct mlx4_en_dev *mdev = priv->mdev;
1611 struct mlx4_en_cq *cq;
1612 struct mlx4_en_tx_ring *tx_ring;
1617 u8 mc_list[16] = {0};
1619 if (priv->port_up) {
1620 en_dbg(DRV, priv, "start port called while port already up\n");
1624 INIT_LIST_HEAD(&priv->mc_list);
1625 INIT_LIST_HEAD(&priv->curr_list);
1626 INIT_LIST_HEAD(&priv->ethtool_list);
1627 memset(&priv->ethtool_rules[0], 0,
1628 sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1630 /* Calculate Rx buf size */
1631 dev->mtu = min(dev->mtu, priv->max_mtu);
1632 mlx4_en_calc_rx_buf(dev);
1633 en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1635 /* Configure rx cq's and rings */
1636 err = mlx4_en_activate_rx_rings(priv);
1638 en_err(priv, "Failed to activate RX rings\n");
1641 for (i = 0; i < priv->rx_ring_num; i++) {
1642 cq = priv->rx_cq[i];
1644 err = mlx4_en_init_affinity_hint(priv, i);
1646 en_err(priv, "Failed preparing IRQ affinity hint\n");
1650 err = mlx4_en_activate_cq(priv, cq, i);
1652 en_err(priv, "Failed activating Rx CQ\n");
1653 mlx4_en_free_affinity_hint(priv, i);
1657 for (j = 0; j < cq->size; j++) {
1658 struct mlx4_cqe *cqe = NULL;
1660 cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1662 cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1665 err = mlx4_en_set_cq_moder(priv, cq);
1667 en_err(priv, "Failed setting cq moderation parameters\n");
1668 mlx4_en_deactivate_cq(priv, cq);
1669 mlx4_en_free_affinity_hint(priv, i);
1672 mlx4_en_arm_cq(priv, cq);
1673 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1678 en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1679 err = mlx4_en_get_qp(priv);
1681 en_err(priv, "Failed getting eth qp\n");
1684 mdev->mac_removed[priv->port] = 0;
1686 priv->counter_index =
1687 mlx4_get_default_counter_index(mdev->dev, priv->port);
1689 err = mlx4_en_config_rss_steer(priv);
1691 en_err(priv, "Failed configuring rss steering\n");
1695 err = mlx4_en_create_drop_qp(priv);
1699 /* Configure tx cq's and rings */
1700 for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1701 u8 num_tx_rings_p_up = t == TX ?
1702 priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1704 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1706 cq = priv->tx_cq[t][i];
1707 err = mlx4_en_activate_cq(priv, cq, i);
1709 en_err(priv, "Failed allocating Tx CQ\n");
1712 err = mlx4_en_set_cq_moder(priv, cq);
1714 en_err(priv, "Failed setting cq moderation parameters\n");
1715 mlx4_en_deactivate_cq(priv, cq);
1719 "Resetting index of collapsed CQ:%d to -1\n", i);
1720 cq->buf->wqe_index = cpu_to_be16(0xffff);
1722 /* Configure ring */
1723 tx_ring = priv->tx_ring[t][i];
1724 err = mlx4_en_activate_tx_ring(priv, tx_ring,
1726 i / num_tx_rings_p_up);
1728 en_err(priv, "Failed allocating Tx ring\n");
1729 mlx4_en_deactivate_cq(priv, cq);
1733 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1734 tx_ring->recycle_ring = NULL;
1736 /* Arm CQ for TX completions */
1737 mlx4_en_arm_cq(priv, cq);
1740 mlx4_en_init_tx_xdp_ring_descs(priv, tx_ring);
1741 mlx4_en_init_recycle_ring(priv, i);
1742 /* XDP TX CQ should never be armed */
1745 /* Set initial ownership of all Tx TXBBs to SW (1) */
1746 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1747 *((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1751 /* Configure port */
1752 err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1753 priv->rx_skb_size + ETH_FCS_LEN,
1754 priv->prof->tx_pause,
1756 priv->prof->rx_pause,
1757 priv->prof->rx_ppp);
1759 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1764 err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1766 en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1767 dev->mtu, priv->port, err);
1771 /* Set default qp number */
1772 err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1774 en_err(priv, "Failed setting default qp numbers\n");
1778 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1779 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1781 en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1788 en_dbg(HW, priv, "Initializing port\n");
1789 err = mlx4_INIT_PORT(mdev->dev, priv->port);
1791 en_err(priv, "Failed Initializing port\n");
1795 /* Set Unicast and VXLAN steering rules */
1796 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1797 mlx4_en_set_rss_steer_rules(priv))
1798 mlx4_warn(mdev, "Failed setting steering rules\n");
1800 /* Attach rx QP to bradcast address */
1801 eth_broadcast_addr(&mc_list[10]);
1802 mc_list[5] = priv->port; /* needed for B0 steering support */
1803 if (mlx4_multicast_attach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1804 priv->port, 0, MLX4_PROT_ETH,
1805 &priv->broadcast_id))
1806 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1808 /* Must redo promiscuous mode setup. */
1809 priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1811 /* Schedule multicast task to populate multicast list */
1812 queue_work(mdev->workqueue, &priv->rx_mode_task);
1814 if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1815 udp_tunnel_get_rx_info(dev);
1817 priv->port_up = true;
1819 /* Process all completions if exist to prevent
1820 * the queues freezing if they are full
1822 for (i = 0; i < priv->rx_ring_num; i++) {
1824 napi_schedule(&priv->rx_cq[i]->napi);
1828 netif_tx_start_all_queues(dev);
1829 netif_device_attach(dev);
1834 if (t == MLX4_EN_NUM_TX_TYPES) {
1836 i = priv->tx_ring_num[t];
1840 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1841 mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1845 i = priv->tx_ring_num[t];
1847 mlx4_en_destroy_drop_qp(priv);
1849 mlx4_en_release_rss_steer(priv);
1851 mlx4_en_put_qp(priv);
1853 while (rx_index--) {
1854 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1855 mlx4_en_free_affinity_hint(priv, rx_index);
1857 for (i = 0; i < priv->rx_ring_num; i++)
1858 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1860 return err; /* need to close devices */
1864 void mlx4_en_stop_port(struct net_device *dev, int detach)
1866 struct mlx4_en_priv *priv = netdev_priv(dev);
1867 struct mlx4_en_dev *mdev = priv->mdev;
1868 struct mlx4_en_mc_list *mclist, *tmp;
1869 struct ethtool_flow_id *flow, *tmp_flow;
1871 u8 mc_list[16] = {0};
1873 if (!priv->port_up) {
1874 en_dbg(DRV, priv, "stop port called while port already down\n");
1879 mlx4_CLOSE_PORT(mdev->dev, priv->port);
1881 /* Synchronize with tx routine */
1882 netif_tx_lock_bh(dev);
1884 netif_device_detach(dev);
1885 netif_tx_stop_all_queues(dev);
1886 netif_tx_unlock_bh(dev);
1888 netif_tx_disable(dev);
1890 spin_lock_bh(&priv->stats_lock);
1891 mlx4_en_fold_software_stats(dev);
1892 /* Set port as not active */
1893 priv->port_up = false;
1894 spin_unlock_bh(&priv->stats_lock);
1896 priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1898 /* Promsicuous mode */
1899 if (mdev->dev->caps.steering_mode ==
1900 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1901 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1902 MLX4_EN_FLAG_MC_PROMISC);
1903 mlx4_flow_steer_promisc_remove(mdev->dev,
1905 MLX4_FS_ALL_DEFAULT);
1906 mlx4_flow_steer_promisc_remove(mdev->dev,
1908 MLX4_FS_MC_DEFAULT);
1909 } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1910 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1912 /* Disable promiscouos mode */
1913 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1916 /* Disable Multicast promisc */
1917 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1918 mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1920 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1924 /* Detach All multicasts */
1925 eth_broadcast_addr(&mc_list[10]);
1926 mc_list[5] = priv->port; /* needed for B0 steering support */
1927 mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1928 MLX4_PROT_ETH, priv->broadcast_id);
1929 list_for_each_entry(mclist, &priv->curr_list, list) {
1930 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1931 mc_list[5] = priv->port;
1932 mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp,
1933 mc_list, MLX4_PROT_ETH, mclist->reg_id);
1934 if (mclist->tunnel_reg_id)
1935 mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1937 mlx4_en_clear_list(dev);
1938 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1939 list_del(&mclist->list);
1943 /* Flush multicast filter */
1944 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1946 /* Remove flow steering rules for the port*/
1947 if (mdev->dev->caps.steering_mode ==
1948 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1950 list_for_each_entry_safe(flow, tmp_flow,
1951 &priv->ethtool_list, list) {
1952 mlx4_flow_detach(mdev->dev, flow->id);
1953 list_del(&flow->list);
1957 mlx4_en_destroy_drop_qp(priv);
1960 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1961 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1962 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1963 mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1968 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1969 for (i = 0; i < priv->tx_ring_num[t]; i++)
1970 mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1972 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1973 mlx4_en_delete_rss_steer_rules(priv);
1976 mlx4_en_release_rss_steer(priv);
1978 /* Unregister Mac address for the port */
1979 mlx4_en_put_qp(priv);
1980 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1981 mdev->mac_removed[priv->port] = 1;
1984 for (i = 0; i < priv->rx_ring_num; i++) {
1985 struct mlx4_en_cq *cq = priv->rx_cq[i];
1987 napi_synchronize(&cq->napi);
1988 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1989 mlx4_en_deactivate_cq(priv, cq);
1991 mlx4_en_free_affinity_hint(priv, i);
1995 static void mlx4_en_restart(struct work_struct *work)
1997 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1999 struct mlx4_en_dev *mdev = priv->mdev;
2000 struct net_device *dev = priv->dev;
2002 en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
2005 mutex_lock(&mdev->state_lock);
2006 if (priv->port_up) {
2007 mlx4_en_stop_port(dev, 1);
2008 if (mlx4_en_start_port(dev))
2009 en_err(priv, "Failed restarting port %d\n", priv->port);
2011 mutex_unlock(&mdev->state_lock);
2015 static void mlx4_en_clear_stats(struct net_device *dev)
2017 struct mlx4_en_priv *priv = netdev_priv(dev);
2018 struct mlx4_en_dev *mdev = priv->mdev;
2019 struct mlx4_en_tx_ring **tx_ring;
2022 if (!mlx4_is_slave(mdev->dev))
2023 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
2024 en_dbg(HW, priv, "Failed dumping statistics\n");
2026 memset(&priv->pstats, 0, sizeof(priv->pstats));
2027 memset(&priv->pkstats, 0, sizeof(priv->pkstats));
2028 memset(&priv->port_stats, 0, sizeof(priv->port_stats));
2029 memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
2030 memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
2031 memset(&priv->rx_priority_flowstats, 0,
2032 sizeof(priv->rx_priority_flowstats));
2033 memset(&priv->tx_priority_flowstats, 0,
2034 sizeof(priv->tx_priority_flowstats));
2035 memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
2037 tx_ring = priv->tx_ring[TX];
2038 for (i = 0; i < priv->tx_ring_num[TX]; i++) {
2039 tx_ring[i]->bytes = 0;
2040 tx_ring[i]->packets = 0;
2041 tx_ring[i]->tx_csum = 0;
2042 tx_ring[i]->tx_dropped = 0;
2043 tx_ring[i]->queue_stopped = 0;
2044 tx_ring[i]->wake_queue = 0;
2045 tx_ring[i]->tso_packets = 0;
2046 tx_ring[i]->xmit_more = 0;
2048 for (i = 0; i < priv->rx_ring_num; i++) {
2049 priv->rx_ring[i]->bytes = 0;
2050 priv->rx_ring[i]->packets = 0;
2051 priv->rx_ring[i]->csum_ok = 0;
2052 priv->rx_ring[i]->csum_none = 0;
2053 priv->rx_ring[i]->csum_complete = 0;
2057 static int mlx4_en_open(struct net_device *dev)
2059 struct mlx4_en_priv *priv = netdev_priv(dev);
2060 struct mlx4_en_dev *mdev = priv->mdev;
2063 mutex_lock(&mdev->state_lock);
2065 if (!mdev->device_up) {
2066 en_err(priv, "Cannot open - device down/disabled\n");
2071 /* Reset HW statistics and SW counters */
2072 mlx4_en_clear_stats(dev);
2074 err = mlx4_en_start_port(dev);
2076 en_err(priv, "Failed starting port:%d\n", priv->port);
2079 mutex_unlock(&mdev->state_lock);
2084 static int mlx4_en_close(struct net_device *dev)
2086 struct mlx4_en_priv *priv = netdev_priv(dev);
2087 struct mlx4_en_dev *mdev = priv->mdev;
2089 en_dbg(IFDOWN, priv, "Close port called\n");
2091 mutex_lock(&mdev->state_lock);
2093 mlx4_en_stop_port(dev, 0);
2094 netif_carrier_off(dev);
2096 mutex_unlock(&mdev->state_lock);
2100 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2104 #ifdef CONFIG_RFS_ACCEL
2105 priv->dev->rx_cpu_rmap = NULL;
2108 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2109 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2110 if (priv->tx_ring[t] && priv->tx_ring[t][i])
2111 mlx4_en_destroy_tx_ring(priv,
2112 &priv->tx_ring[t][i]);
2113 if (priv->tx_cq[t] && priv->tx_cq[t][i])
2114 mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2116 kfree(priv->tx_ring[t]);
2117 kfree(priv->tx_cq[t]);
2120 for (i = 0; i < priv->rx_ring_num; i++) {
2121 if (priv->rx_ring[i])
2122 mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2123 priv->prof->rx_ring_size, priv->stride);
2125 mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2130 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2132 struct mlx4_en_port_profile *prof = priv->prof;
2136 /* Create tx Rings */
2137 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2138 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2139 node = cpu_to_node(i % num_online_cpus());
2140 if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2141 prof->tx_ring_size, i, t, node))
2144 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2146 TXBB_SIZE, node, i))
2151 /* Create rx Rings */
2152 for (i = 0; i < priv->rx_ring_num; i++) {
2153 node = cpu_to_node(i % num_online_cpus());
2154 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2155 prof->rx_ring_size, i, RX, node))
2158 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2159 prof->rx_ring_size, priv->stride,
2165 #ifdef CONFIG_RFS_ACCEL
2166 priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2172 en_err(priv, "Failed to allocate NIC resources\n");
2173 for (i = 0; i < priv->rx_ring_num; i++) {
2174 if (priv->rx_ring[i])
2175 mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2179 mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2181 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2182 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2183 if (priv->tx_ring[t][i])
2184 mlx4_en_destroy_tx_ring(priv,
2185 &priv->tx_ring[t][i]);
2186 if (priv->tx_cq[t][i])
2187 mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2194 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2195 struct mlx4_en_priv *src,
2196 struct mlx4_en_port_profile *prof)
2200 memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2201 sizeof(dst->hwtstamp_config));
2202 dst->num_tx_rings_p_up = prof->num_tx_rings_p_up;
2203 dst->rx_ring_num = prof->rx_ring_num;
2204 dst->flags = prof->flags;
2205 dst->mdev = src->mdev;
2206 dst->port = src->port;
2207 dst->dev = src->dev;
2209 dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2210 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2212 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2213 dst->tx_ring_num[t] = prof->tx_ring_num[t];
2214 if (!dst->tx_ring_num[t])
2217 dst->tx_ring[t] = kcalloc(MAX_TX_RINGS,
2218 sizeof(struct mlx4_en_tx_ring *),
2220 if (!dst->tx_ring[t])
2223 dst->tx_cq[t] = kcalloc(MAX_TX_RINGS,
2224 sizeof(struct mlx4_en_cq *),
2226 if (!dst->tx_cq[t]) {
2227 kfree(dst->tx_ring[t]);
2236 kfree(dst->tx_ring[t]);
2237 kfree(dst->tx_cq[t]);
2242 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2243 struct mlx4_en_priv *src)
2246 memcpy(dst->rx_ring, src->rx_ring,
2247 sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2248 memcpy(dst->rx_cq, src->rx_cq,
2249 sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2250 memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2251 sizeof(dst->hwtstamp_config));
2252 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2253 dst->tx_ring_num[t] = src->tx_ring_num[t];
2254 dst->tx_ring[t] = src->tx_ring[t];
2255 dst->tx_cq[t] = src->tx_cq[t];
2257 dst->num_tx_rings_p_up = src->num_tx_rings_p_up;
2258 dst->rx_ring_num = src->rx_ring_num;
2259 memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2262 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2263 struct mlx4_en_priv *tmp,
2264 struct mlx4_en_port_profile *prof,
2265 bool carry_xdp_prog)
2267 struct bpf_prog *xdp_prog;
2270 mlx4_en_copy_priv(tmp, priv, prof);
2272 if (mlx4_en_alloc_resources(tmp)) {
2274 "%s: Resource allocation failed, using previous configuration\n",
2276 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2277 kfree(tmp->tx_ring[t]);
2278 kfree(tmp->tx_cq[t]);
2283 /* All rx_rings has the same xdp_prog. Pick the first one. */
2284 xdp_prog = rcu_dereference_protected(
2285 priv->rx_ring[0]->xdp_prog,
2286 lockdep_is_held(&priv->mdev->state_lock));
2288 if (xdp_prog && carry_xdp_prog) {
2289 xdp_prog = bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2290 if (IS_ERR(xdp_prog)) {
2291 mlx4_en_free_resources(tmp);
2292 return PTR_ERR(xdp_prog);
2294 for (i = 0; i < tmp->rx_ring_num; i++)
2295 rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2302 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2303 struct mlx4_en_priv *tmp)
2305 mlx4_en_free_resources(priv);
2306 mlx4_en_update_priv(priv, tmp);
2309 void mlx4_en_destroy_netdev(struct net_device *dev)
2311 struct mlx4_en_priv *priv = netdev_priv(dev);
2312 struct mlx4_en_dev *mdev = priv->mdev;
2314 en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2316 /* Unregister device - this will close the port if it was up */
2317 if (priv->registered) {
2318 devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2320 unregister_netdev(dev);
2323 if (priv->allocated)
2324 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2326 cancel_delayed_work(&priv->stats_task);
2327 cancel_delayed_work(&priv->service_task);
2328 /* flush any pending task for this netdev */
2329 flush_workqueue(mdev->workqueue);
2331 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2332 mlx4_en_remove_timestamp(mdev);
2334 /* Detach the netdev so tasks would not attempt to access it */
2335 mutex_lock(&mdev->state_lock);
2336 mdev->pndev[priv->port] = NULL;
2337 mdev->upper[priv->port] = NULL;
2339 #ifdef CONFIG_RFS_ACCEL
2340 mlx4_en_cleanup_filters(priv);
2343 mlx4_en_free_resources(priv);
2344 mutex_unlock(&mdev->state_lock);
2349 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2351 struct mlx4_en_priv *priv = netdev_priv(dev);
2353 if (mtu > MLX4_EN_MAX_XDP_MTU) {
2354 en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2355 mtu, MLX4_EN_MAX_XDP_MTU);
2362 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2364 struct mlx4_en_priv *priv = netdev_priv(dev);
2365 struct mlx4_en_dev *mdev = priv->mdev;
2368 en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2371 if (priv->tx_ring_num[TX_XDP] &&
2372 !mlx4_en_check_xdp_mtu(dev, new_mtu))
2377 if (netif_running(dev)) {
2378 mutex_lock(&mdev->state_lock);
2379 if (!mdev->device_up) {
2380 /* NIC is probably restarting - let watchdog task reset
2382 en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2384 mlx4_en_stop_port(dev, 1);
2385 err = mlx4_en_start_port(dev);
2387 en_err(priv, "Failed restarting port:%d\n",
2389 queue_work(mdev->workqueue, &priv->watchdog_task);
2392 mutex_unlock(&mdev->state_lock);
2397 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2399 struct mlx4_en_priv *priv = netdev_priv(dev);
2400 struct mlx4_en_dev *mdev = priv->mdev;
2401 struct hwtstamp_config config;
2403 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2406 /* reserved for future extensions */
2410 /* device doesn't support time stamping */
2411 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2414 /* TX HW timestamp */
2415 switch (config.tx_type) {
2416 case HWTSTAMP_TX_OFF:
2417 case HWTSTAMP_TX_ON:
2423 /* RX HW timestamp */
2424 switch (config.rx_filter) {
2425 case HWTSTAMP_FILTER_NONE:
2427 case HWTSTAMP_FILTER_ALL:
2428 case HWTSTAMP_FILTER_SOME:
2429 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2430 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2431 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2432 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2433 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2434 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2435 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2436 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2437 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2438 case HWTSTAMP_FILTER_PTP_V2_EVENT:
2439 case HWTSTAMP_FILTER_PTP_V2_SYNC:
2440 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2441 case HWTSTAMP_FILTER_NTP_ALL:
2442 config.rx_filter = HWTSTAMP_FILTER_ALL;
2448 if (mlx4_en_reset_config(dev, config, dev->features)) {
2449 config.tx_type = HWTSTAMP_TX_OFF;
2450 config.rx_filter = HWTSTAMP_FILTER_NONE;
2453 return copy_to_user(ifr->ifr_data, &config,
2454 sizeof(config)) ? -EFAULT : 0;
2457 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2459 struct mlx4_en_priv *priv = netdev_priv(dev);
2461 return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2462 sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2465 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2469 return mlx4_en_hwtstamp_set(dev, ifr);
2471 return mlx4_en_hwtstamp_get(dev, ifr);
2477 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2478 netdev_features_t features)
2480 struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2481 struct mlx4_en_dev *mdev = en_priv->mdev;
2483 /* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2484 * enable/disable make sure S-TAG flag is always in same state as
2487 if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2488 !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2489 features |= NETIF_F_HW_VLAN_STAG_RX;
2491 features &= ~NETIF_F_HW_VLAN_STAG_RX;
2496 static int mlx4_en_set_features(struct net_device *netdev,
2497 netdev_features_t features)
2499 struct mlx4_en_priv *priv = netdev_priv(netdev);
2503 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2504 en_info(priv, "Turn %s RX-FCS\n",
2505 (features & NETIF_F_RXFCS) ? "ON" : "OFF");
2509 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2510 u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2512 en_info(priv, "Turn %s RX-ALL\n",
2513 ignore_fcs_value ? "ON" : "OFF");
2514 ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2515 priv->port, ignore_fcs_value);
2520 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2521 en_info(priv, "Turn %s RX vlan strip offload\n",
2522 (features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2526 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2527 en_info(priv, "Turn %s TX vlan strip offload\n",
2528 (features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2530 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2531 en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2532 (features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2534 if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2535 en_info(priv, "Turn %s loopback\n",
2536 (features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2537 mlx4_en_update_loopback_state(netdev, features);
2541 ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2550 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2552 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2553 struct mlx4_en_dev *mdev = en_priv->mdev;
2555 return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac);
2558 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2561 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2562 struct mlx4_en_dev *mdev = en_priv->mdev;
2564 return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2568 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2571 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2572 struct mlx4_en_dev *mdev = en_priv->mdev;
2574 return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2578 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2580 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2581 struct mlx4_en_dev *mdev = en_priv->mdev;
2583 return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2586 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2588 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2589 struct mlx4_en_dev *mdev = en_priv->mdev;
2591 return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2594 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2596 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2597 struct mlx4_en_dev *mdev = en_priv->mdev;
2599 return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2602 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2603 struct ifla_vf_stats *vf_stats)
2605 struct mlx4_en_priv *en_priv = netdev_priv(dev);
2606 struct mlx4_en_dev *mdev = en_priv->mdev;
2608 return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2611 #define PORT_ID_BYTE_LEN 8
2612 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2613 struct netdev_phys_item_id *ppid)
2615 struct mlx4_en_priv *priv = netdev_priv(dev);
2616 struct mlx4_dev *mdev = priv->mdev->dev;
2618 u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2623 ppid->id_len = sizeof(phys_port_id);
2624 for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2625 ppid->id[i] = phys_port_id & 0xff;
2631 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2634 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2637 ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2641 ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2642 VXLAN_STEER_BY_OUTER_MAC, 1);
2645 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2650 priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2652 NETIF_F_TSO | NETIF_F_TSO6 |
2653 NETIF_F_GSO_UDP_TUNNEL |
2654 NETIF_F_GSO_UDP_TUNNEL_CSUM |
2655 NETIF_F_GSO_PARTIAL;
2658 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2661 struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2663 /* unset offloads */
2664 priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2666 NETIF_F_TSO | NETIF_F_TSO6 |
2667 NETIF_F_GSO_UDP_TUNNEL |
2668 NETIF_F_GSO_UDP_TUNNEL_CSUM |
2669 NETIF_F_GSO_PARTIAL);
2671 ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2672 VXLAN_STEER_BY_OUTER_MAC, 0);
2674 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2676 priv->vxlan_port = 0;
2679 static void mlx4_en_add_vxlan_port(struct net_device *dev,
2680 struct udp_tunnel_info *ti)
2682 struct mlx4_en_priv *priv = netdev_priv(dev);
2683 __be16 port = ti->port;
2684 __be16 current_port;
2686 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2689 if (ti->sa_family != AF_INET)
2692 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2695 current_port = priv->vxlan_port;
2696 if (current_port && current_port != port) {
2697 en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2698 ntohs(current_port), ntohs(port));
2702 priv->vxlan_port = port;
2703 queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2706 static void mlx4_en_del_vxlan_port(struct net_device *dev,
2707 struct udp_tunnel_info *ti)
2709 struct mlx4_en_priv *priv = netdev_priv(dev);
2710 __be16 port = ti->port;
2711 __be16 current_port;
2713 if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2716 if (ti->sa_family != AF_INET)
2719 if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2722 current_port = priv->vxlan_port;
2723 if (current_port != port) {
2724 en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2728 queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2731 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2732 struct net_device *dev,
2733 netdev_features_t features)
2735 features = vlan_features_check(skb, features);
2736 features = vxlan_features_check(skb, features);
2738 /* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2739 * support inner IPv6 checksums and segmentation so we need to
2740 * strip that feature if this is an IPv6 encapsulated frame.
2742 if (skb->encapsulation &&
2743 (skb->ip_summed == CHECKSUM_PARTIAL)) {
2744 struct mlx4_en_priv *priv = netdev_priv(dev);
2746 if (!priv->vxlan_port ||
2747 (ip_hdr(skb)->version != 4) ||
2748 (udp_hdr(skb)->dest != priv->vxlan_port))
2749 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2755 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2757 struct mlx4_en_priv *priv = netdev_priv(dev);
2758 struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2759 struct mlx4_update_qp_params params;
2762 if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2765 /* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2766 if (maxrate >> 12) {
2767 params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2768 params.rate_val = maxrate / 1000;
2769 } else if (maxrate) {
2770 params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2771 params.rate_val = maxrate;
2772 } else { /* zero serves to revoke the QP rate-limitation */
2773 params.rate_unit = 0;
2774 params.rate_val = 0;
2777 err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2782 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2784 struct mlx4_en_priv *priv = netdev_priv(dev);
2785 struct mlx4_en_dev *mdev = priv->mdev;
2786 struct mlx4_en_port_profile new_prof;
2787 struct bpf_prog *old_prog;
2788 struct mlx4_en_priv *tmp;
2795 xdp_ring_num = prog ? priv->rx_ring_num : 0;
2797 /* No need to reconfigure buffers when simply swapping the
2798 * program for a new one.
2800 if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2802 prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2804 return PTR_ERR(prog);
2806 mutex_lock(&mdev->state_lock);
2807 for (i = 0; i < priv->rx_ring_num; i++) {
2808 old_prog = rcu_dereference_protected(
2809 priv->rx_ring[i]->xdp_prog,
2810 lockdep_is_held(&mdev->state_lock));
2811 rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2813 bpf_prog_put(old_prog);
2815 mutex_unlock(&mdev->state_lock);
2819 if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2822 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2827 prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2829 err = PTR_ERR(prog);
2834 mutex_lock(&mdev->state_lock);
2835 memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2836 new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2838 if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2840 new_prof.tx_ring_num[TX] =
2841 MAX_TX_RINGS - ALIGN(xdp_ring_num, priv->prof->num_up);
2842 en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2845 err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2848 bpf_prog_sub(prog, priv->rx_ring_num - 1);
2852 if (priv->port_up) {
2854 mlx4_en_stop_port(dev, 1);
2857 mlx4_en_safe_replace_resources(priv, tmp);
2859 netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2861 for (i = 0; i < priv->rx_ring_num; i++) {
2862 old_prog = rcu_dereference_protected(
2863 priv->rx_ring[i]->xdp_prog,
2864 lockdep_is_held(&mdev->state_lock));
2865 rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2867 bpf_prog_put(old_prog);
2871 err = mlx4_en_start_port(dev);
2873 en_err(priv, "Failed starting port %d for XDP change\n",
2875 queue_work(mdev->workqueue, &priv->watchdog_task);
2880 mutex_unlock(&mdev->state_lock);
2886 static u32 mlx4_xdp_query(struct net_device *dev)
2888 struct mlx4_en_priv *priv = netdev_priv(dev);
2889 struct mlx4_en_dev *mdev = priv->mdev;
2890 const struct bpf_prog *xdp_prog;
2893 if (!priv->tx_ring_num[TX_XDP])
2896 mutex_lock(&mdev->state_lock);
2897 xdp_prog = rcu_dereference_protected(
2898 priv->rx_ring[0]->xdp_prog,
2899 lockdep_is_held(&mdev->state_lock));
2901 prog_id = xdp_prog->aux->id;
2902 mutex_unlock(&mdev->state_lock);
2907 static int mlx4_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2909 switch (xdp->command) {
2910 case XDP_SETUP_PROG:
2911 return mlx4_xdp_set(dev, xdp->prog);
2912 case XDP_QUERY_PROG:
2913 xdp->prog_id = mlx4_xdp_query(dev);
2920 static const struct net_device_ops mlx4_netdev_ops = {
2921 .ndo_open = mlx4_en_open,
2922 .ndo_stop = mlx4_en_close,
2923 .ndo_start_xmit = mlx4_en_xmit,
2924 .ndo_select_queue = mlx4_en_select_queue,
2925 .ndo_get_stats64 = mlx4_en_get_stats64,
2926 .ndo_set_rx_mode = mlx4_en_set_rx_mode,
2927 .ndo_set_mac_address = mlx4_en_set_mac,
2928 .ndo_validate_addr = eth_validate_addr,
2929 .ndo_change_mtu = mlx4_en_change_mtu,
2930 .ndo_do_ioctl = mlx4_en_ioctl,
2931 .ndo_tx_timeout = mlx4_en_tx_timeout,
2932 .ndo_vlan_rx_add_vid = mlx4_en_vlan_rx_add_vid,
2933 .ndo_vlan_rx_kill_vid = mlx4_en_vlan_rx_kill_vid,
2934 .ndo_set_features = mlx4_en_set_features,
2935 .ndo_fix_features = mlx4_en_fix_features,
2936 .ndo_setup_tc = __mlx4_en_setup_tc,
2937 #ifdef CONFIG_RFS_ACCEL
2938 .ndo_rx_flow_steer = mlx4_en_filter_rfs,
2940 .ndo_get_phys_port_id = mlx4_en_get_phys_port_id,
2941 .ndo_udp_tunnel_add = mlx4_en_add_vxlan_port,
2942 .ndo_udp_tunnel_del = mlx4_en_del_vxlan_port,
2943 .ndo_features_check = mlx4_en_features_check,
2944 .ndo_set_tx_maxrate = mlx4_en_set_tx_maxrate,
2945 .ndo_bpf = mlx4_xdp,
2948 static const struct net_device_ops mlx4_netdev_ops_master = {
2949 .ndo_open = mlx4_en_open,
2950 .ndo_stop = mlx4_en_close,
2951 .ndo_start_xmit = mlx4_en_xmit,
2952 .ndo_select_queue = mlx4_en_select_queue,
2953 .ndo_get_stats64 = mlx4_en_get_stats64,
2954 .ndo_set_rx_mode = mlx4_en_set_rx_mode,
2955 .ndo_set_mac_address = mlx4_en_set_mac,
2956 .ndo_validate_addr = eth_validate_addr,
2957 .ndo_change_mtu = mlx4_en_change_mtu,
2958 .ndo_tx_timeout = mlx4_en_tx_timeout,
2959 .ndo_vlan_rx_add_vid = mlx4_en_vlan_rx_add_vid,
2960 .ndo_vlan_rx_kill_vid = mlx4_en_vlan_rx_kill_vid,
2961 .ndo_set_vf_mac = mlx4_en_set_vf_mac,
2962 .ndo_set_vf_vlan = mlx4_en_set_vf_vlan,
2963 .ndo_set_vf_rate = mlx4_en_set_vf_rate,
2964 .ndo_set_vf_spoofchk = mlx4_en_set_vf_spoofchk,
2965 .ndo_set_vf_link_state = mlx4_en_set_vf_link_state,
2966 .ndo_get_vf_stats = mlx4_en_get_vf_stats,
2967 .ndo_get_vf_config = mlx4_en_get_vf_config,
2968 .ndo_set_features = mlx4_en_set_features,
2969 .ndo_fix_features = mlx4_en_fix_features,
2970 .ndo_setup_tc = __mlx4_en_setup_tc,
2971 #ifdef CONFIG_RFS_ACCEL
2972 .ndo_rx_flow_steer = mlx4_en_filter_rfs,
2974 .ndo_get_phys_port_id = mlx4_en_get_phys_port_id,
2975 .ndo_udp_tunnel_add = mlx4_en_add_vxlan_port,
2976 .ndo_udp_tunnel_del = mlx4_en_del_vxlan_port,
2977 .ndo_features_check = mlx4_en_features_check,
2978 .ndo_set_tx_maxrate = mlx4_en_set_tx_maxrate,
2979 .ndo_bpf = mlx4_xdp,
2982 struct mlx4_en_bond {
2983 struct work_struct work;
2984 struct mlx4_en_priv *priv;
2986 struct mlx4_port_map port_map;
2989 static void mlx4_en_bond_work(struct work_struct *work)
2991 struct mlx4_en_bond *bond = container_of(work,
2992 struct mlx4_en_bond,
2995 struct mlx4_dev *dev = bond->priv->mdev->dev;
2997 if (bond->is_bonded) {
2998 if (!mlx4_is_bonded(dev)) {
2999 err = mlx4_bond(dev);
3001 en_err(bond->priv, "Fail to bond device\n");
3004 err = mlx4_port_map_set(dev, &bond->port_map);
3006 en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
3007 bond->port_map.port1,
3008 bond->port_map.port2,
3011 } else if (mlx4_is_bonded(dev)) {
3012 err = mlx4_unbond(dev);
3014 en_err(bond->priv, "Fail to unbond device\n");
3016 dev_put(bond->priv->dev);
3020 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
3021 u8 v2p_p1, u8 v2p_p2)
3023 struct mlx4_en_bond *bond = NULL;
3025 bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
3029 INIT_WORK(&bond->work, mlx4_en_bond_work);
3031 bond->is_bonded = is_bonded;
3032 bond->port_map.port1 = v2p_p1;
3033 bond->port_map.port2 = v2p_p2;
3034 dev_hold(priv->dev);
3035 queue_work(priv->mdev->workqueue, &bond->work);
3039 int mlx4_en_netdev_event(struct notifier_block *this,
3040 unsigned long event, void *ptr)
3042 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3044 struct mlx4_en_dev *mdev;
3045 struct mlx4_dev *dev;
3046 int i, num_eth_ports = 0;
3047 bool do_bond = true;
3048 struct mlx4_en_priv *priv;
3052 if (!net_eq(dev_net(ndev), &init_net))
3055 mdev = container_of(this, struct mlx4_en_dev, nb);
3058 /* Go into this mode only when two network devices set on two ports
3059 * of the same mlx4 device are slaves of the same bonding master
3061 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
3063 if (!port && (mdev->pndev[i] == ndev))
3065 mdev->upper[i] = mdev->pndev[i] ?
3066 netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
3067 /* condition not met: network device is a slave */
3068 if (!mdev->upper[i])
3070 if (num_eth_ports < 2)
3072 /* condition not met: same master */
3073 if (mdev->upper[i] != mdev->upper[i-1])
3076 /* condition not met: 2 salves */
3077 do_bond = (num_eth_ports == 2) ? do_bond : false;
3079 /* handle only events that come with enough info */
3080 if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
3083 priv = netdev_priv(ndev);
3085 struct netdev_notifier_bonding_info *notifier_info = ptr;
3086 struct netdev_bonding_info *bonding_info =
3087 ¬ifier_info->bonding_info;
3089 /* required mode 1, 2 or 4 */
3090 if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3091 (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
3092 (bonding_info->master.bond_mode != BOND_MODE_8023AD))
3095 /* require exactly 2 slaves */
3096 if (bonding_info->master.num_slaves != 2)
3101 if (bonding_info->master.bond_mode ==
3102 BOND_MODE_ACTIVEBACKUP) {
3103 /* in active-backup mode virtual ports are
3104 * mapped to the physical port of the active
3106 if (bonding_info->slave.state ==
3107 BOND_STATE_BACKUP) {
3115 } else { /* BOND_STATE_ACTIVE */
3124 } else { /* Active-Active */
3125 /* in active-active mode a virtual port is
3126 * mapped to the native physical port if and only
3127 * if the physical port is up */
3128 __s8 link = bonding_info->slave.link;
3134 if ((link == BOND_LINK_UP) ||
3135 (link == BOND_LINK_FAIL)) {
3140 } else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3150 mlx4_en_queue_bond_work(priv, do_bond,
3151 v2p_port1, v2p_port2);
3156 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3157 struct mlx4_en_stats_bitmap *stats_bitmap,
3158 u8 rx_ppp, u8 rx_pause,
3159 u8 tx_ppp, u8 tx_pause)
3161 int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3163 if (!mlx4_is_slave(dev) &&
3164 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3165 mutex_lock(&stats_bitmap->mutex);
3166 bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3169 bitmap_set(stats_bitmap->bitmap, last_i,
3170 NUM_FLOW_PRIORITY_STATS_RX);
3171 last_i += NUM_FLOW_PRIORITY_STATS_RX;
3173 if (rx_pause && !(rx_ppp))
3174 bitmap_set(stats_bitmap->bitmap, last_i,
3176 last_i += NUM_FLOW_STATS_RX;
3179 bitmap_set(stats_bitmap->bitmap, last_i,
3180 NUM_FLOW_PRIORITY_STATS_TX);
3181 last_i += NUM_FLOW_PRIORITY_STATS_TX;
3183 if (tx_pause && !(tx_ppp))
3184 bitmap_set(stats_bitmap->bitmap, last_i,
3186 last_i += NUM_FLOW_STATS_TX;
3188 mutex_unlock(&stats_bitmap->mutex);
3192 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3193 struct mlx4_en_stats_bitmap *stats_bitmap,
3194 u8 rx_ppp, u8 rx_pause,
3195 u8 tx_ppp, u8 tx_pause)
3199 mutex_init(&stats_bitmap->mutex);
3200 bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3202 if (mlx4_is_slave(dev)) {
3203 bitmap_set(stats_bitmap->bitmap, last_i +
3204 MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3205 bitmap_set(stats_bitmap->bitmap, last_i +
3206 MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3207 bitmap_set(stats_bitmap->bitmap, last_i +
3208 MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3209 bitmap_set(stats_bitmap->bitmap, last_i +
3210 MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3211 bitmap_set(stats_bitmap->bitmap, last_i +
3212 MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3213 bitmap_set(stats_bitmap->bitmap, last_i +
3214 MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3216 bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3218 last_i += NUM_MAIN_STATS;
3220 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3221 last_i += NUM_PORT_STATS;
3223 if (mlx4_is_master(dev))
3224 bitmap_set(stats_bitmap->bitmap, last_i,
3226 last_i += NUM_PF_STATS;
3228 mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3231 last_i += NUM_FLOW_STATS;
3233 if (!mlx4_is_slave(dev))
3234 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3235 last_i += NUM_PKT_STATS;
3237 bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3238 last_i += NUM_XDP_STATS;
3240 if (!mlx4_is_slave(dev))
3241 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PHY_STATS);
3242 last_i += NUM_PHY_STATS;
3245 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3246 struct mlx4_en_port_profile *prof)
3248 struct net_device *dev;
3249 struct mlx4_en_priv *priv;
3253 dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3254 MAX_TX_RINGS, MAX_RX_RINGS);
3258 netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3259 netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3261 SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3262 dev->dev_port = port - 1;
3265 * Initialize driver private data
3268 priv = netdev_priv(dev);
3269 memset(priv, 0, sizeof(struct mlx4_en_priv));
3270 priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3271 spin_lock_init(&priv->stats_lock);
3272 INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3273 INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
3274 INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3275 INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3276 INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3277 INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
3278 INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
3279 #ifdef CONFIG_RFS_ACCEL
3280 INIT_LIST_HEAD(&priv->filters);
3281 spin_lock_init(&priv->filters_lock);
3286 priv->ddev = &mdev->pdev->dev;
3289 priv->port_up = false;
3290 priv->flags = prof->flags;
3291 priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3292 priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3293 MLX4_WQE_CTRL_SOLICITED);
3294 priv->num_tx_rings_p_up = mdev->profile.max_num_tx_rings_p_up;
3295 priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3296 netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3298 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3299 priv->tx_ring_num[t] = prof->tx_ring_num[t];
3300 if (!priv->tx_ring_num[t])
3303 priv->tx_ring[t] = kcalloc(MAX_TX_RINGS,
3304 sizeof(struct mlx4_en_tx_ring *),
3306 if (!priv->tx_ring[t]) {
3310 priv->tx_cq[t] = kcalloc(MAX_TX_RINGS,
3311 sizeof(struct mlx4_en_cq *),
3313 if (!priv->tx_cq[t]) {
3318 priv->rx_ring_num = prof->rx_ring_num;
3319 priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3320 priv->cqe_size = mdev->dev->caps.cqe_size;
3321 priv->mac_index = -1;
3322 priv->msg_enable = MLX4_EN_MSG_LEVEL;
3323 #ifdef CONFIG_MLX4_EN_DCB
3324 if (!mlx4_is_slave(priv->mdev->dev)) {
3327 for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; ++prio) {
3328 priv->ets.prio_tc[prio] = prio;
3329 priv->ets.tc_tsa[prio] = IEEE_8021QAZ_TSA_VENDOR;
3332 priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3333 DCB_CAP_DCBX_VER_IEEE;
3334 priv->flags |= MLX4_EN_DCB_ENABLED;
3335 priv->cee_config.pfc_state = false;
3337 for (i = 0; i < MLX4_EN_NUM_UP_HIGH; i++)
3338 priv->cee_config.dcb_pfc[i] = pfc_disabled;
3340 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3341 dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3343 en_info(priv, "enabling only PFC DCB ops\n");
3344 dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3349 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3350 INIT_HLIST_HEAD(&priv->mac_hash[i]);
3352 /* Query for default mac and max mtu */
3353 priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3355 if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3356 MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3357 priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3359 /* Set default MAC */
3360 dev->addr_len = ETH_ALEN;
3361 mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3362 if (!is_valid_ether_addr(dev->dev_addr)) {
3363 en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
3364 priv->port, dev->dev_addr);
3367 } else if (mlx4_is_slave(priv->mdev->dev) &&
3368 (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3369 /* Random MAC was assigned in mlx4_slave_cap
3370 * in mlx4_core module
3372 dev->addr_assign_type |= NET_ADDR_RANDOM;
3373 en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3376 memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3378 priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3379 DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3380 err = mlx4_en_alloc_resources(priv);
3384 /* Initialize time stamping config */
3385 priv->hwtstamp_config.flags = 0;
3386 priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3387 priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3389 /* Allocate page for receive rings */
3390 err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3393 en_err(priv, "Failed to allocate page for rx qps\n");
3396 priv->allocated = 1;
3399 * Initialize netdev entry points
3401 if (mlx4_is_master(priv->mdev->dev))
3402 dev->netdev_ops = &mlx4_netdev_ops_master;
3404 dev->netdev_ops = &mlx4_netdev_ops;
3405 dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3406 netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3407 netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3409 dev->ethtool_ops = &mlx4_en_ethtool_ops;
3412 * Set driver features
3414 dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3415 if (mdev->LSO_support)
3416 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3418 dev->vlan_features = dev->hw_features;
3420 dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3421 dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3422 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3423 NETIF_F_HW_VLAN_CTAG_FILTER;
3424 dev->hw_features |= NETIF_F_LOOPBACK |
3425 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3427 if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3428 dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3429 NETIF_F_HW_VLAN_STAG_FILTER;
3430 dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3433 if (mlx4_is_slave(mdev->dev)) {
3434 bool vlan_offload_disabled;
3437 err = get_phv_bit(mdev->dev, port, &phv);
3439 dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3440 priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3442 err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3443 &vlan_offload_disabled);
3444 if (!err && vlan_offload_disabled) {
3445 dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3446 NETIF_F_HW_VLAN_CTAG_RX |
3447 NETIF_F_HW_VLAN_STAG_TX |
3448 NETIF_F_HW_VLAN_STAG_RX);
3449 dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3450 NETIF_F_HW_VLAN_CTAG_RX |
3451 NETIF_F_HW_VLAN_STAG_TX |
3452 NETIF_F_HW_VLAN_STAG_RX);
3455 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3456 !(mdev->dev->caps.flags2 &
3457 MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3458 dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3461 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3462 dev->hw_features |= NETIF_F_RXFCS;
3464 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3465 dev->hw_features |= NETIF_F_RXALL;
3467 if (mdev->dev->caps.steering_mode ==
3468 MLX4_STEERING_MODE_DEVICE_MANAGED &&
3469 mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3470 dev->hw_features |= NETIF_F_NTUPLE;
3472 if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3473 dev->priv_flags |= IFF_UNICAST_FLT;
3475 /* Setting a default hash function value */
3476 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3477 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3478 } else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3479 priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3482 "No RSS hash capabilities exposed, using Toeplitz\n");
3483 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3486 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3487 dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3488 NETIF_F_GSO_UDP_TUNNEL_CSUM |
3489 NETIF_F_GSO_PARTIAL;
3490 dev->features |= NETIF_F_GSO_UDP_TUNNEL |
3491 NETIF_F_GSO_UDP_TUNNEL_CSUM |
3492 NETIF_F_GSO_PARTIAL;
3493 dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3496 /* MTU range: 68 - hw-specific max */
3497 dev->min_mtu = ETH_MIN_MTU;
3498 dev->max_mtu = priv->max_mtu;
3500 mdev->pndev[port] = dev;
3501 mdev->upper[port] = NULL;
3503 netif_carrier_off(dev);
3504 mlx4_en_set_default_moderation(priv);
3506 en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3507 en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3509 mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3511 /* Configure port */
3512 mlx4_en_calc_rx_buf(dev);
3513 err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3514 priv->rx_skb_size + ETH_FCS_LEN,
3515 prof->tx_pause, prof->tx_ppp,
3516 prof->rx_pause, prof->rx_ppp);
3518 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3523 if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3524 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3526 en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3533 en_warn(priv, "Initializing port\n");
3534 err = mlx4_INIT_PORT(mdev->dev, priv->port);
3536 en_err(priv, "Failed Initializing port\n");
3539 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3541 /* Initialize time stamp mechanism */
3542 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3543 mlx4_en_init_timestamp(mdev);
3545 queue_delayed_work(mdev->workqueue, &priv->service_task,
3546 SERVICE_TASK_DELAY);
3548 mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3549 mdev->profile.prof[priv->port].rx_ppp,
3550 mdev->profile.prof[priv->port].rx_pause,
3551 mdev->profile.prof[priv->port].tx_ppp,
3552 mdev->profile.prof[priv->port].tx_pause);
3554 err = register_netdev(dev);
3556 en_err(priv, "Netdev registration failed for port %d\n", port);
3560 priv->registered = 1;
3561 devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3567 mlx4_en_destroy_netdev(dev);
3571 int mlx4_en_reset_config(struct net_device *dev,
3572 struct hwtstamp_config ts_config,
3573 netdev_features_t features)
3575 struct mlx4_en_priv *priv = netdev_priv(dev);
3576 struct mlx4_en_dev *mdev = priv->mdev;
3577 struct mlx4_en_port_profile new_prof;
3578 struct mlx4_en_priv *tmp;
3582 if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3583 priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3584 !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3585 !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3586 return 0; /* Nothing to change */
3588 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3589 (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3590 (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3591 en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3595 tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3599 mutex_lock(&mdev->state_lock);
3601 memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3602 memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3604 err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3608 if (priv->port_up) {
3610 mlx4_en_stop_port(dev, 1);
3613 mlx4_en_safe_replace_resources(priv, tmp);
3615 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3616 if (features & NETIF_F_HW_VLAN_CTAG_RX)
3617 dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3619 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3620 } else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3621 /* RX time-stamping is OFF, update the RX vlan offload
3622 * to the latest wanted state
3624 if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3625 dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3627 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3630 if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3631 if (features & NETIF_F_RXFCS)
3632 dev->features |= NETIF_F_RXFCS;
3634 dev->features &= ~NETIF_F_RXFCS;
3637 /* RX vlan offload and RX time-stamping can't co-exist !
3638 * Regardless of the caller's choice,
3639 * Turn Off RX vlan offload in case of time-stamping is ON
3641 if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3642 if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3643 en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3644 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3648 err = mlx4_en_start_port(dev);
3650 en_err(priv, "Failed starting port\n");
3654 mutex_unlock(&mdev->state_lock);
3657 netdev_features_change(dev);