drivers/net/ethernet/mellanox/mlx4/en_netdev.c

   1 /*
   2  * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  *
  32  */
  33
  34 #include <linux/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>
  41 #include <net/ip.h>
  42 #include <net/busy_poll.h>
  43 #include <net/vxlan.h>
  44 #include <net/devlink.h>
  45
  46 #include <linux/mlx4/driver.h>
  47 #include <linux/mlx4/device.h>
  48 #include <linux/mlx4/cmd.h>
  49 #include <linux/mlx4/cq.h>
  50
  51 #include "mlx4_en.h"
  52 #include "en_port.h"
  53
  54 #define MLX4_EN_MAX_XDP_MTU ((int)(PAGE_SIZE - ETH_HLEN - (2 * VLAN_HLEN) - \
  55                                    XDP_PACKET_HEADROOM))
  56
  57 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
  58 {
  59         struct mlx4_en_priv *priv = netdev_priv(dev);
  60         int i;
  61         unsigned int offset = 0;
  62
  63         if (up && up != MLX4_EN_NUM_UP_HIGH)
  64                 return -EINVAL;
  65
  66         netdev_set_num_tc(dev, up);
  67         netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
  68         /* Partition Tx queues evenly amongst UP's */
  69         for (i = 0; i < up; i++) {
  70                 netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
  71                 offset += priv->num_tx_rings_p_up;
  72         }
  73
  74 #ifdef CONFIG_MLX4_EN_DCB
  75         if (!mlx4_is_slave(priv->mdev->dev)) {
  76                 if (up) {
  77                         if (priv->dcbx_cap)
  78                                 priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
  79                 } else {
  80                         priv->flags &= ~MLX4_EN_FLAG_DCB_ENABLED;
  81                         priv->cee_config.pfc_state = false;
  82                 }
  83         }
  84 #endif /* CONFIG_MLX4_EN_DCB */
  85
  86         return 0;
  87 }
  88
  89 int mlx4_en_alloc_tx_queue_per_tc(struct net_device *dev, u8 tc)
  90 {
  91         struct mlx4_en_priv *priv = netdev_priv(dev);
  92         struct mlx4_en_dev *mdev = priv->mdev;
  93         struct mlx4_en_port_profile new_prof;
  94         struct mlx4_en_priv *tmp;
  95         int port_up = 0;
  96         int err = 0;
  97
  98         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
  99         if (!tmp)
 100                 return -ENOMEM;
 101
 102         mutex_lock(&mdev->state_lock);
 103         memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
 104         new_prof.num_up = (tc == 0) ? MLX4_EN_NUM_UP_LOW :
 105                                       MLX4_EN_NUM_UP_HIGH;
 106         new_prof.tx_ring_num[TX] = new_prof.num_tx_rings_p_up *
 107                                    new_prof.num_up;
 108         err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
 109         if (err)
 110                 goto out;
 111
 112         if (priv->port_up) {
 113                 port_up = 1;
 114                 mlx4_en_stop_port(dev, 1);
 115         }
 116
 117         mlx4_en_safe_replace_resources(priv, tmp);
 118         if (port_up) {
 119                 err = mlx4_en_start_port(dev);
 120                 if (err) {
 121                         en_err(priv, "Failed starting port for setup TC\n");
 122                         goto out;
 123                 }
 124         }
 125
 126         err = mlx4_en_setup_tc(dev, tc);
 127 out:
 128         mutex_unlock(&mdev->state_lock);
 129         kfree(tmp);
 130         return err;
 131 }
 132
 133 static int __mlx4_en_setup_tc(struct net_device *dev, enum tc_setup_type type,
 134                               void *type_data)
 135 {
 136         struct tc_mqprio_qopt *mqprio = type_data;
 137
 138         if (type != TC_SETUP_MQPRIO)
 139                 return -EOPNOTSUPP;
 140
 141         if (mqprio->num_tc && mqprio->num_tc != MLX4_EN_NUM_UP_HIGH)
 142                 return -EINVAL;
 143
 144         mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
 145
 146         return mlx4_en_alloc_tx_queue_per_tc(dev, mqprio->num_tc);
 147 }
 148
 149 #ifdef CONFIG_RFS_ACCEL
 150
 151 struct mlx4_en_filter {
 152         struct list_head next;
 153         struct work_struct work;
 154
 155         u8     ip_proto;
 156         __be32 src_ip;
 157         __be32 dst_ip;
 158         __be16 src_port;
 159         __be16 dst_port;
 160
 161         int rxq_index;
 162         struct mlx4_en_priv *priv;
 163         u32 flow_id;                    /* RFS infrastructure id */
 164         int id;                         /* mlx4_en driver id */
 165         u64 reg_id;                     /* Flow steering API id */
 166         u8 activated;                   /* Used to prevent expiry before filter
 167                                          * is attached
 168                                          */
 169         struct hlist_node filter_chain;
 170 };
 171
 172 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
 173
 174 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
 175 {
 176         switch (ip_proto) {
 177         case IPPROTO_UDP:
 178                 return MLX4_NET_TRANS_RULE_ID_UDP;
 179         case IPPROTO_TCP:
 180                 return MLX4_NET_TRANS_RULE_ID_TCP;
 181         default:
 182                 return MLX4_NET_TRANS_RULE_NUM;
 183         }
 184 };
 185
 186 /* Must not acquire state_lock, as its corresponding work_sync
 187  * is done under it.
 188  */
 189 static void mlx4_en_filter_work(struct work_struct *work)
 190 {
 191         struct mlx4_en_filter *filter = container_of(work,
 192                                                      struct mlx4_en_filter,
 193                                                      work);
 194         struct mlx4_en_priv *priv = filter->priv;
 195         struct mlx4_spec_list spec_tcp_udp = {
 196                 .id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
 197                 {
 198                         .tcp_udp = {
 199                                 .dst_port = filter->dst_port,
 200                                 .dst_port_msk = (__force __be16)-1,
 201                                 .src_port = filter->src_port,
 202                                 .src_port_msk = (__force __be16)-1,
 203                         },
 204                 },
 205         };
 206         struct mlx4_spec_list spec_ip = {
 207                 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
 208                 {
 209                         .ipv4 = {
 210                                 .dst_ip = filter->dst_ip,
 211                                 .dst_ip_msk = (__force __be32)-1,
 212                                 .src_ip = filter->src_ip,
 213                                 .src_ip_msk = (__force __be32)-1,
 214                         },
 215                 },
 216         };
 217         struct mlx4_spec_list spec_eth = {
 218                 .id = MLX4_NET_TRANS_RULE_ID_ETH,
 219         };
 220         struct mlx4_net_trans_rule rule = {
 221                 .list = LIST_HEAD_INIT(rule.list),
 222                 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
 223                 .exclusive = 1,
 224                 .allow_loopback = 1,
 225                 .promisc_mode = MLX4_FS_REGULAR,
 226                 .port = priv->port,
 227                 .priority = MLX4_DOMAIN_RFS,
 228         };
 229         int rc;
 230         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 231
 232         if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
 233                 en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
 234                         filter->ip_proto);
 235                 goto ignore;
 236         }
 237         list_add_tail(&spec_eth.list, &rule.list);
 238         list_add_tail(&spec_ip.list, &rule.list);
 239         list_add_tail(&spec_tcp_udp.list, &rule.list);
 240
 241         rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
 242         memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
 243         memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 244
 245         filter->activated = 0;
 246
 247         if (filter->reg_id) {
 248                 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 249                 if (rc && rc != -ENOENT)
 250                         en_err(priv, "Error detaching flow. rc = %d\n", rc);
 251         }
 252
 253         rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
 254         if (rc)
 255                 en_err(priv, "Error attaching flow. err = %d\n", rc);
 256
 257 ignore:
 258         mlx4_en_filter_rfs_expire(priv);
 259
 260         filter->activated = 1;
 261 }
 262
 263 static inline struct hlist_head *
 264 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 265                    __be16 src_port, __be16 dst_port)
 266 {
 267         unsigned long l;
 268         int bucket_idx;
 269
 270         l = (__force unsigned long)src_port |
 271             ((__force unsigned long)dst_port << 2);
 272         l ^= (__force unsigned long)(src_ip ^ dst_ip);
 273
 274         bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
 275
 276         return &priv->filter_hash[bucket_idx];
 277 }
 278
 279 static struct mlx4_en_filter *
 280 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
 281                      __be32 dst_ip, u8 ip_proto, __be16 src_port,
 282                      __be16 dst_port, u32 flow_id)
 283 {
 284         struct mlx4_en_filter *filter = NULL;
 285
 286         filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
 287         if (!filter)
 288                 return NULL;
 289
 290         filter->priv = priv;
 291         filter->rxq_index = rxq_index;
 292         INIT_WORK(&filter->work, mlx4_en_filter_work);
 293
 294         filter->src_ip = src_ip;
 295         filter->dst_ip = dst_ip;
 296         filter->ip_proto = ip_proto;
 297         filter->src_port = src_port;
 298         filter->dst_port = dst_port;
 299
 300         filter->flow_id = flow_id;
 301
 302         filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
 303
 304         list_add_tail(&filter->next, &priv->filters);
 305         hlist_add_head(&filter->filter_chain,
 306                        filter_hash_bucket(priv, src_ip, dst_ip, src_port,
 307                                           dst_port));
 308
 309         return filter;
 310 }
 311
 312 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
 313 {
 314         struct mlx4_en_priv *priv = filter->priv;
 315         int rc;
 316
 317         list_del(&filter->next);
 318
 319         rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 320         if (rc && rc != -ENOENT)
 321                 en_err(priv, "Error detaching flow. rc = %d\n", rc);
 322
 323         kfree(filter);
 324 }
 325
 326 static inline struct mlx4_en_filter *
 327 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 328                     u8 ip_proto, __be16 src_port, __be16 dst_port)
 329 {
 330         struct mlx4_en_filter *filter;
 331         struct mlx4_en_filter *ret = NULL;
 332
 333         hlist_for_each_entry(filter,
 334                              filter_hash_bucket(priv, src_ip, dst_ip,
 335                                                 src_port, dst_port),
 336                              filter_chain) {
 337                 if (filter->src_ip == src_ip &&
 338                     filter->dst_ip == dst_ip &&
 339                     filter->ip_proto == ip_proto &&
 340                     filter->src_port == src_port &&
 341                     filter->dst_port == dst_port) {
 342                         ret = filter;
 343                         break;
 344                 }
 345         }
 346
 347         return ret;
 348 }
 349
 350 static int
 351 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
 352                    u16 rxq_index, u32 flow_id)
 353 {
 354         struct mlx4_en_priv *priv = netdev_priv(net_dev);
 355         struct mlx4_en_filter *filter;
 356         const struct iphdr *ip;
 357         const __be16 *ports;
 358         u8 ip_proto;
 359         __be32 src_ip;
 360         __be32 dst_ip;
 361         __be16 src_port;
 362         __be16 dst_port;
 363         int nhoff = skb_network_offset(skb);
 364         int ret = 0;
 365
 366         if (skb->encapsulation)
 367                 return -EPROTONOSUPPORT;
 368
 369         if (skb->protocol != htons(ETH_P_IP))
 370                 return -EPROTONOSUPPORT;
 371
 372         ip = (const struct iphdr *)(skb->data + nhoff);
 373         if (ip_is_fragment(ip))
 374                 return -EPROTONOSUPPORT;
 375
 376         if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
 377                 return -EPROTONOSUPPORT;
 378         ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
 379
 380         ip_proto = ip->protocol;
 381         src_ip = ip->saddr;
 382         dst_ip = ip->daddr;
 383         src_port = ports[0];
 384         dst_port = ports[1];
 385
 386         spin_lock_bh(&priv->filters_lock);
 387         filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
 388                                      src_port, dst_port);
 389         if (filter) {
 390                 if (filter->rxq_index == rxq_index)
 391                         goto out;
 392
 393                 filter->rxq_index = rxq_index;
 394         } else {
 395                 filter = mlx4_en_filter_alloc(priv, rxq_index,
 396                                               src_ip, dst_ip, ip_proto,
 397                                               src_port, dst_port, flow_id);
 398                 if (!filter) {
 399                         ret = -ENOMEM;
 400                         goto err;
 401                 }
 402         }
 403
 404         queue_work(priv->mdev->workqueue, &filter->work);
 405
 406 out:
 407         ret = filter->id;
 408 err:
 409         spin_unlock_bh(&priv->filters_lock);
 410
 411         return ret;
 412 }
 413
 414 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
 415 {
 416         struct mlx4_en_filter *filter, *tmp;
 417         LIST_HEAD(del_list);
 418
 419         spin_lock_bh(&priv->filters_lock);
 420         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 421                 list_move(&filter->next, &del_list);
 422                 hlist_del(&filter->filter_chain);
 423         }
 424         spin_unlock_bh(&priv->filters_lock);
 425
 426         list_for_each_entry_safe(filter, tmp, &del_list, next) {
 427                 cancel_work_sync(&filter->work);
 428                 mlx4_en_filter_free(filter);
 429         }
 430 }
 431
 432 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
 433 {
 434         struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
 435         LIST_HEAD(del_list);
 436         int i = 0;
 437
 438         spin_lock_bh(&priv->filters_lock);
 439         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 440                 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
 441                         break;
 442
 443                 if (filter->activated &&
 444                     !work_pending(&filter->work) &&
 445                     rps_may_expire_flow(priv->dev,
 446                                         filter->rxq_index, filter->flow_id,
 447                                         filter->id)) {
 448                         list_move(&filter->next, &del_list);
 449                         hlist_del(&filter->filter_chain);
 450                 } else
 451                         last_filter = filter;
 452
 453                 i++;
 454         }
 455
 456         if (last_filter && (&last_filter->next != priv->filters.next))
 457                 list_move(&priv->filters, &last_filter->next);
 458
 459         spin_unlock_bh(&priv->filters_lock);
 460
 461         list_for_each_entry_safe(filter, tmp, &del_list, next)
 462                 mlx4_en_filter_free(filter);
 463 }
 464 #endif
 465
 466 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
 467                                    __be16 proto, u16 vid)
 468 {
 469         struct mlx4_en_priv *priv = netdev_priv(dev);
 470         struct mlx4_en_dev *mdev = priv->mdev;
 471         int err;
 472         int idx;
 473
 474         en_dbg(HW, priv, "adding VLAN:%d\n", vid);
 475
 476         set_bit(vid, priv->active_vlans);
 477
 478         /* Add VID to port VLAN filter */
 479         mutex_lock(&mdev->state_lock);
 480         if (mdev->device_up && priv->port_up) {
 481                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 482                 if (err) {
 483                         en_err(priv, "Failed configuring VLAN filter\n");
 484                         goto out;
 485                 }
 486         }
 487         err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
 488         if (err)
 489                 en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
 490
 491 out:
 492         mutex_unlock(&mdev->state_lock);
 493         return err;
 494 }
 495
 496 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
 497                                     __be16 proto, u16 vid)
 498 {
 499         struct mlx4_en_priv *priv = netdev_priv(dev);
 500         struct mlx4_en_dev *mdev = priv->mdev;
 501         int err = 0;
 502
 503         en_dbg(HW, priv, "Killing VID:%d\n", vid);
 504
 505         clear_bit(vid, priv->active_vlans);
 506
 507         /* Remove VID from port VLAN filter */
 508         mutex_lock(&mdev->state_lock);
 509         mlx4_unregister_vlan(mdev->dev, priv->port, vid);
 510
 511         if (mdev->device_up && priv->port_up) {
 512                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 513                 if (err)
 514                         en_err(priv, "Failed configuring VLAN filter\n");
 515         }
 516         mutex_unlock(&mdev->state_lock);
 517
 518         return err;
 519 }
 520
 521 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
 522 {
 523         int i;
 524         for (i = ETH_ALEN - 1; i >= 0; --i) {
 525                 dst_mac[i] = src_mac & 0xff;
 526                 src_mac >>= 8;
 527         }
 528         memset(&dst_mac[ETH_ALEN], 0, 2);
 529 }
 530
 531
 532 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
 533                                     int qpn, u64 *reg_id)
 534 {
 535         int err;
 536
 537         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
 538             priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
 539                 return 0; /* do nothing */
 540
 541         err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
 542                                     MLX4_DOMAIN_NIC, reg_id);
 543         if (err) {
 544                 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
 545                 return err;
 546         }
 547         en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
 548         return 0;
 549 }
 550
 551
 552 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
 553                                 unsigned char *mac, int *qpn, u64 *reg_id)
 554 {
 555         struct mlx4_en_dev *mdev = priv->mdev;
 556         struct mlx4_dev *dev = mdev->dev;
 557         int err;
 558
 559         switch (dev->caps.steering_mode) {
 560         case MLX4_STEERING_MODE_B0: {
 561                 struct mlx4_qp qp;
 562                 u8 gid[16] = {0};
 563
 564                 qp.qpn = *qpn;
 565                 memcpy(&gid[10], mac, ETH_ALEN);
 566                 gid[5] = priv->port;
 567
 568                 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
 569                 break;
 570         }
 571         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 572                 struct mlx4_spec_list spec_eth = { {NULL} };
 573                 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 574
 575                 struct mlx4_net_trans_rule rule = {
 576                         .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 577                         .exclusive = 0,
 578                         .allow_loopback = 1,
 579                         .promisc_mode = MLX4_FS_REGULAR,
 580                         .priority = MLX4_DOMAIN_NIC,
 581                 };
 582
 583                 rule.port = priv->port;
 584                 rule.qpn = *qpn;
 585                 INIT_LIST_HEAD(&rule.list);
 586
 587                 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
 588                 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
 589                 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 590                 list_add_tail(&spec_eth.list, &rule.list);
 591
 592                 err = mlx4_flow_attach(dev, &rule, reg_id);
 593                 break;
 594         }
 595         default:
 596                 return -EINVAL;
 597         }
 598         if (err)
 599                 en_warn(priv, "Failed Attaching Unicast\n");
 600
 601         return err;
 602 }
 603
 604 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
 605                                      unsigned char *mac, int qpn, u64 reg_id)
 606 {
 607         struct mlx4_en_dev *mdev = priv->mdev;
 608         struct mlx4_dev *dev = mdev->dev;
 609
 610         switch (dev->caps.steering_mode) {
 611         case MLX4_STEERING_MODE_B0: {
 612                 struct mlx4_qp qp;
 613                 u8 gid[16] = {0};
 614
 615                 qp.qpn = qpn;
 616                 memcpy(&gid[10], mac, ETH_ALEN);
 617                 gid[5] = priv->port;
 618
 619                 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
 620                 break;
 621         }
 622         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 623                 mlx4_flow_detach(dev, reg_id);
 624                 break;
 625         }
 626         default:
 627                 en_err(priv, "Invalid steering mode.\n");
 628         }
 629 }
 630
 631 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
 632 {
 633         struct mlx4_en_dev *mdev = priv->mdev;
 634         struct mlx4_dev *dev = mdev->dev;
 635         int index = 0;
 636         int err = 0;
 637         int *qpn = &priv->base_qpn;
 638         u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 639
 640         en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
 641                priv->dev->dev_addr);
 642         index = mlx4_register_mac(dev, priv->port, mac);
 643         if (index < 0) {
 644                 err = index;
 645                 en_err(priv, "Failed adding MAC: %pM\n",
 646                        priv->dev->dev_addr);
 647                 return err;
 648         }
 649
 650         en_info(priv, "Steering Mode %d\n", dev->caps.steering_mode);
 651
 652         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 653                 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
 654                 *qpn = base_qpn + index;
 655                 return 0;
 656         }
 657
 658         err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP,
 659                                     MLX4_RES_USAGE_DRIVER);
 660         en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
 661         if (err) {
 662                 en_err(priv, "Failed to reserve qp for mac registration\n");
 663                 mlx4_unregister_mac(dev, priv->port, mac);
 664                 return err;
 665         }
 666
 667         return 0;
 668 }
 669
 670 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
 671 {
 672         struct mlx4_en_dev *mdev = priv->mdev;
 673         struct mlx4_dev *dev = mdev->dev;
 674         int qpn = priv->base_qpn;
 675
 676         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 677                 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 678                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 679                        priv->dev->dev_addr);
 680                 mlx4_unregister_mac(dev, priv->port, mac);
 681         } else {
 682                 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
 683                        priv->port, qpn);
 684                 mlx4_qp_release_range(dev, qpn, 1);
 685                 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
 686         }
 687 }
 688
 689 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
 690                                unsigned char *new_mac, unsigned char *prev_mac)
 691 {
 692         struct mlx4_en_dev *mdev = priv->mdev;
 693         struct mlx4_dev *dev = mdev->dev;
 694         int err = 0;
 695         u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
 696
 697         if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
 698                 struct hlist_head *bucket;
 699                 unsigned int mac_hash;
 700                 struct mlx4_mac_entry *entry;
 701                 struct hlist_node *tmp;
 702                 u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
 703
 704                 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
 705                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 706                         if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
 707                                 mlx4_en_uc_steer_release(priv, entry->mac,
 708                                                          qpn, entry->reg_id);
 709                                 mlx4_unregister_mac(dev, priv->port,
 710                                                     prev_mac_u64);
 711                                 hlist_del_rcu(&entry->hlist);
 712                                 synchronize_rcu();
 713                                 memcpy(entry->mac, new_mac, ETH_ALEN);
 714                                 entry->reg_id = 0;
 715                                 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
 716                                 hlist_add_head_rcu(&entry->hlist,
 717                                                    &priv->mac_hash[mac_hash]);
 718                                 mlx4_register_mac(dev, priv->port, new_mac_u64);
 719                                 err = mlx4_en_uc_steer_add(priv, new_mac,
 720                                                            &qpn,
 721                                                            &entry->reg_id);
 722                                 if (err)
 723                                         return err;
 724                                 if (priv->tunnel_reg_id) {
 725                                         mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 726                                         priv->tunnel_reg_id = 0;
 727                                 }
 728                                 err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
 729                                                                &priv->tunnel_reg_id);
 730                                 return err;
 731                         }
 732                 }
 733                 return -EINVAL;
 734         }
 735
 736         return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
 737 }
 738
 739 static void mlx4_en_update_user_mac(struct mlx4_en_priv *priv,
 740                                     unsigned char new_mac[ETH_ALEN + 2])
 741 {
 742         struct mlx4_en_dev *mdev = priv->mdev;
 743         int err;
 744
 745         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_USER_MAC_EN))
 746                 return;
 747
 748         err = mlx4_SET_PORT_user_mac(mdev->dev, priv->port, new_mac);
 749         if (err)
 750                 en_err(priv, "Failed to pass user MAC(%pM) to Firmware for port %d, with error %d\n",
 751                        new_mac, priv->port, err);
 752 }
 753
 754 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
 755                               unsigned char new_mac[ETH_ALEN + 2])
 756 {
 757         int err = 0;
 758
 759         if (priv->port_up) {
 760                 /* Remove old MAC and insert the new one */
 761                 err = mlx4_en_replace_mac(priv, priv->base_qpn,
 762                                           new_mac, priv->current_mac);
 763                 if (err)
 764                         en_err(priv, "Failed changing HW MAC address\n");
 765         } else
 766                 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
 767
 768         if (!err)
 769                 memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
 770
 771         return err;
 772 }
 773
 774 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
 775 {
 776         struct mlx4_en_priv *priv = netdev_priv(dev);
 777         struct mlx4_en_dev *mdev = priv->mdev;
 778         struct sockaddr *saddr = addr;
 779         unsigned char new_mac[ETH_ALEN + 2];
 780         int err;
 781
 782         if (!is_valid_ether_addr(saddr->sa_data))
 783                 return -EADDRNOTAVAIL;
 784
 785         mutex_lock(&mdev->state_lock);
 786         memcpy(new_mac, saddr->sa_data, ETH_ALEN);
 787         err = mlx4_en_do_set_mac(priv, new_mac);
 788         if (err)
 789                 goto out;
 790
 791         memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
 792         mlx4_en_update_user_mac(priv, new_mac);
 793 out:
 794         mutex_unlock(&mdev->state_lock);
 795
 796         return err;
 797 }
 798
 799 static void mlx4_en_clear_list(struct net_device *dev)
 800 {
 801         struct mlx4_en_priv *priv = netdev_priv(dev);
 802         struct mlx4_en_mc_list *tmp, *mc_to_del;
 803
 804         list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
 805                 list_del(&mc_to_del->list);
 806                 kfree(mc_to_del);
 807         }
 808 }
 809
 810 static void mlx4_en_cache_mclist(struct net_device *dev)
 811 {
 812         struct mlx4_en_priv *priv = netdev_priv(dev);
 813         struct netdev_hw_addr *ha;
 814         struct mlx4_en_mc_list *tmp;
 815
 816         mlx4_en_clear_list(dev);
 817         netdev_for_each_mc_addr(ha, dev) {
 818                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 819                 if (!tmp) {
 820                         mlx4_en_clear_list(dev);
 821                         return;
 822                 }
 823                 memcpy(tmp->addr, ha->addr, ETH_ALEN);
 824                 list_add_tail(&tmp->list, &priv->mc_list);
 825         }
 826 }
 827
 828 static void update_mclist_flags(struct mlx4_en_priv *priv,
 829                                 struct list_head *dst,
 830                                 struct list_head *src)
 831 {
 832         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 833         bool found;
 834
 835         /* Find all the entries that should be removed from dst,
 836          * These are the entries that are not found in src
 837          */
 838         list_for_each_entry(dst_tmp, dst, list) {
 839                 found = false;
 840                 list_for_each_entry(src_tmp, src, list) {
 841                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 842                                 found = true;
 843                                 break;
 844                         }
 845                 }
 846                 if (!found)
 847                         dst_tmp->action = MCLIST_REM;
 848         }
 849
 850         /* Add entries that exist in src but not in dst
 851          * mark them as need to add
 852          */
 853         list_for_each_entry(src_tmp, src, list) {
 854                 found = false;
 855                 list_for_each_entry(dst_tmp, dst, list) {
 856                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 857                                 dst_tmp->action = MCLIST_NONE;
 858                                 found = true;
 859                                 break;
 860                         }
 861                 }
 862                 if (!found) {
 863                         new_mc = kmemdup(src_tmp,
 864                                          sizeof(struct mlx4_en_mc_list),
 865                                          GFP_KERNEL);
 866                         if (!new_mc)
 867                                 return;
 868
 869                         new_mc->action = MCLIST_ADD;
 870                         list_add_tail(&new_mc->list, dst);
 871                 }
 872         }
 873 }
 874
 875 static void mlx4_en_set_rx_mode(struct net_device *dev)
 876 {
 877         struct mlx4_en_priv *priv = netdev_priv(dev);
 878
 879         if (!priv->port_up)
 880                 return;
 881
 882         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 883 }
 884
 885 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 886                                      struct mlx4_en_dev *mdev)
 887 {
 888         int err = 0;
 889
 890         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 891                 if (netif_msg_rx_status(priv))
 892                         en_warn(priv, "Entering promiscuous mode\n");
 893                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 894
 895                 /* Enable promiscouos mode */
 896                 switch (mdev->dev->caps.steering_mode) {
 897                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 898                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 899                                                           priv->port,
 900                                                           priv->base_qpn,
 901                                                           MLX4_FS_ALL_DEFAULT);
 902                         if (err)
 903                                 en_err(priv, "Failed enabling promiscuous mode\n");
 904                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 905                         break;
 906
 907                 case MLX4_STEERING_MODE_B0:
 908                         err = mlx4_unicast_promisc_add(mdev->dev,
 909                                                        priv->base_qpn,
 910                                                        priv->port);
 911                         if (err)
 912                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 913
 914                         /* Add the default qp number as multicast
 915                          * promisc
 916                          */
 917                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 918                                 err = mlx4_multicast_promisc_add(mdev->dev,
 919                                                                  priv->base_qpn,
 920                                                                  priv->port);
 921                                 if (err)
 922                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 923                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 924                         }
 925                         break;
 926
 927                 case MLX4_STEERING_MODE_A0:
 928                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 929                                                      priv->port,
 930                                                      priv->base_qpn,
 931                                                      1);
 932                         if (err)
 933                                 en_err(priv, "Failed enabling promiscuous mode\n");
 934                         break;
 935                 }
 936
 937                 /* Disable port multicast filter (unconditionally) */
 938                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 939                                           0, MLX4_MCAST_DISABLE);
 940                 if (err)
 941                         en_err(priv, "Failed disabling multicast filter\n");
 942         }
 943 }
 944
 945 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 946                                        struct mlx4_en_dev *mdev)
 947 {
 948         int err = 0;
 949
 950         if (netif_msg_rx_status(priv))
 951                 en_warn(priv, "Leaving promiscuous mode\n");
 952         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 953
 954         /* Disable promiscouos mode */
 955         switch (mdev->dev->caps.steering_mode) {
 956         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 957                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 958                                                      priv->port,
 959                                                      MLX4_FS_ALL_DEFAULT);
 960                 if (err)
 961                         en_err(priv, "Failed disabling promiscuous mode\n");
 962                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 963                 break;
 964
 965         case MLX4_STEERING_MODE_B0:
 966                 err = mlx4_unicast_promisc_remove(mdev->dev,
 967                                                   priv->base_qpn,
 968                                                   priv->port);
 969                 if (err)
 970                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 971                 /* Disable Multicast promisc */
 972                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 973                         err = mlx4_multicast_promisc_remove(mdev->dev,
 974                                                             priv->base_qpn,
 975                                                             priv->port);
 976                         if (err)
 977                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 978                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 979                 }
 980                 break;
 981
 982         case MLX4_STEERING_MODE_A0:
 983                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 984                                              priv->port,
 985                                              priv->base_qpn, 0);
 986                 if (err)
 987                         en_err(priv, "Failed disabling promiscuous mode\n");
 988                 break;
 989         }
 990 }
 991
 992 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
 993                                  struct net_device *dev,
 994                                  struct mlx4_en_dev *mdev)
 995 {
 996         struct mlx4_en_mc_list *mclist, *tmp;
 997         u64 mcast_addr = 0;
 998         u8 mc_list[16] = {0};
 999         int err = 0;
1000
1001         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
1002         if (dev->flags & IFF_ALLMULTI) {
1003                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1004                                           0, MLX4_MCAST_DISABLE);
1005                 if (err)
1006                         en_err(priv, "Failed disabling multicast filter\n");
1007
1008                 /* Add the default qp number as multicast promisc */
1009                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1010                         switch (mdev->dev->caps.steering_mode) {
1011                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
1012                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
1013                                                                   priv->port,
1014                                                                   priv->base_qpn,
1015                                                                   MLX4_FS_MC_DEFAULT);
1016                                 break;
1017
1018                         case MLX4_STEERING_MODE_B0:
1019                                 err = mlx4_multicast_promisc_add(mdev->dev,
1020                                                                  priv->base_qpn,
1021                                                                  priv->port);
1022                                 break;
1023
1024                         case MLX4_STEERING_MODE_A0:
1025                                 break;
1026                         }
1027                         if (err)
1028                                 en_err(priv, "Failed entering multicast promisc mode\n");
1029                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1030                 }
1031         } else {
1032                 /* Disable Multicast promisc */
1033                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1034                         switch (mdev->dev->caps.steering_mode) {
1035                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
1036                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
1037                                                                      priv->port,
1038                                                                      MLX4_FS_MC_DEFAULT);
1039                                 break;
1040
1041                         case MLX4_STEERING_MODE_B0:
1042                                 err = mlx4_multicast_promisc_remove(mdev->dev,
1043                                                                     priv->base_qpn,
1044                                                                     priv->port);
1045                                 break;
1046
1047                         case MLX4_STEERING_MODE_A0:
1048                                 break;
1049                         }
1050                         if (err)
1051                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
1052                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1053                 }
1054
1055                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1056                                           0, MLX4_MCAST_DISABLE);
1057                 if (err)
1058                         en_err(priv, "Failed disabling multicast filter\n");
1059
1060                 /* Flush mcast filter and init it with broadcast address */
1061                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1062                                     1, MLX4_MCAST_CONFIG);
1063
1064                 /* Update multicast list - we cache all addresses so they won't
1065                  * change while HW is updated holding the command semaphor */
1066                 netif_addr_lock_bh(dev);
1067                 mlx4_en_cache_mclist(dev);
1068                 netif_addr_unlock_bh(dev);
1069                 list_for_each_entry(mclist, &priv->mc_list, list) {
1070                         mcast_addr = mlx4_mac_to_u64(mclist->addr);
1071                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1072                                             mcast_addr, 0, MLX4_MCAST_CONFIG);
1073                 }
1074                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1075                                           0, MLX4_MCAST_ENABLE);
1076                 if (err)
1077                         en_err(priv, "Failed enabling multicast filter\n");
1078
1079                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1080                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1081                         if (mclist->action == MCLIST_REM) {
1082                                 /* detach this address and delete from list */
1083                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1084                                 mc_list[5] = priv->port;
1085                                 err = mlx4_multicast_detach(mdev->dev,
1086                                                             priv->rss_map.indir_qp,
1087                                                             mc_list,
1088                                                             MLX4_PROT_ETH,
1089                                                             mclist->reg_id);
1090                                 if (err)
1091                                         en_err(priv, "Fail to detach multicast address\n");
1092
1093                                 if (mclist->tunnel_reg_id) {
1094                                         err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1095                                         if (err)
1096                                                 en_err(priv, "Failed to detach multicast address\n");
1097                                 }
1098
1099                                 /* remove from list */
1100                                 list_del(&mclist->list);
1101                                 kfree(mclist);
1102                         } else if (mclist->action == MCLIST_ADD) {
1103                                 /* attach the address */
1104                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1105                                 /* needed for B0 steering support */
1106                                 mc_list[5] = priv->port;
1107                                 err = mlx4_multicast_attach(mdev->dev,
1108                                                             priv->rss_map.indir_qp,
1109                                                             mc_list,
1110                                                             priv->port, 0,
1111                                                             MLX4_PROT_ETH,
1112                                                             &mclist->reg_id);
1113                                 if (err)
1114                                         en_err(priv, "Fail to attach multicast address\n");
1115
1116                                 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1117                                                                &mclist->tunnel_reg_id);
1118                                 if (err)
1119                                         en_err(priv, "Failed to attach multicast address\n");
1120                         }
1121                 }
1122         }
1123 }
1124
1125 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1126                                  struct net_device *dev,
1127                                  struct mlx4_en_dev *mdev)
1128 {
1129         struct netdev_hw_addr *ha;
1130         struct mlx4_mac_entry *entry;
1131         struct hlist_node *tmp;
1132         bool found;
1133         u64 mac;
1134         int err = 0;
1135         struct hlist_head *bucket;
1136         unsigned int i;
1137         int removed = 0;
1138         u32 prev_flags;
1139
1140         /* Note that we do not need to protect our mac_hash traversal with rcu,
1141          * since all modification code is protected by mdev->state_lock
1142          */
1143
1144         /* find what to remove */
1145         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1146                 bucket = &priv->mac_hash[i];
1147                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1148                         found = false;
1149                         netdev_for_each_uc_addr(ha, dev) {
1150                                 if (ether_addr_equal_64bits(entry->mac,
1151                                                             ha->addr)) {
1152                                         found = true;
1153                                         break;
1154                                 }
1155                         }
1156
1157                         /* MAC address of the port is not in uc list */
1158                         if (ether_addr_equal_64bits(entry->mac,
1159                                                     priv->current_mac))
1160                                 found = true;
1161
1162                         if (!found) {
1163                                 mac = mlx4_mac_to_u64(entry->mac);
1164                                 mlx4_en_uc_steer_release(priv, entry->mac,
1165                                                          priv->base_qpn,
1166                                                          entry->reg_id);
1167                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1168
1169                                 hlist_del_rcu(&entry->hlist);
1170                                 kfree_rcu(entry, rcu);
1171                                 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1172                                        entry->mac, priv->port);
1173                                 ++removed;
1174                         }
1175                 }
1176         }
1177
1178         /* if we didn't remove anything, there is no use in trying to add
1179          * again once we are in a forced promisc mode state
1180          */
1181         if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1182                 return;
1183
1184         prev_flags = priv->flags;
1185         priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1186
1187         /* find what to add */
1188         netdev_for_each_uc_addr(ha, dev) {
1189                 found = false;
1190                 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1191                 hlist_for_each_entry(entry, bucket, hlist) {
1192                         if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1193                                 found = true;
1194                                 break;
1195                         }
1196                 }
1197
1198                 if (!found) {
1199                         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1200                         if (!entry) {
1201                                 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1202                                        ha->addr, priv->port);
1203                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1204                                 break;
1205                         }
1206                         mac = mlx4_mac_to_u64(ha->addr);
1207                         memcpy(entry->mac, ha->addr, ETH_ALEN);
1208                         err = mlx4_register_mac(mdev->dev, priv->port, mac);
1209                         if (err < 0) {
1210                                 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1211                                        ha->addr, priv->port, err);
1212                                 kfree(entry);
1213                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1214                                 break;
1215                         }
1216                         err = mlx4_en_uc_steer_add(priv, ha->addr,
1217                                                    &priv->base_qpn,
1218                                                    &entry->reg_id);
1219                         if (err) {
1220                                 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1221                                        ha->addr, priv->port, err);
1222                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1223                                 kfree(entry);
1224                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1225                                 break;
1226                         } else {
1227                                 unsigned int mac_hash;
1228                                 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1229                                        ha->addr, priv->port);
1230                                 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1231                                 bucket = &priv->mac_hash[mac_hash];
1232                                 hlist_add_head_rcu(&entry->hlist, bucket);
1233                         }
1234                 }
1235         }
1236
1237         if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1238                 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1239                         priv->port);
1240         } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1241                 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1242                         priv->port);
1243         }
1244 }
1245
1246 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1247 {
1248         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1249                                                  rx_mode_task);
1250         struct mlx4_en_dev *mdev = priv->mdev;
1251         struct net_device *dev = priv->dev;
1252
1253         mutex_lock(&mdev->state_lock);
1254         if (!mdev->device_up) {
1255                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1256                 goto out;
1257         }
1258         if (!priv->port_up) {
1259                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1260                 goto out;
1261         }
1262
1263         if (!netif_carrier_ok(dev)) {
1264                 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1265                         if (priv->port_state.link_state) {
1266                                 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1267                                 netif_carrier_on(dev);
1268                                 en_dbg(LINK, priv, "Link Up\n");
1269                         }
1270                 }
1271         }
1272
1273         if (dev->priv_flags & IFF_UNICAST_FLT)
1274                 mlx4_en_do_uc_filter(priv, dev, mdev);
1275
1276         /* Promsicuous mode: disable all filters */
1277         if ((dev->flags & IFF_PROMISC) ||
1278             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1279                 mlx4_en_set_promisc_mode(priv, mdev);
1280                 goto out;
1281         }
1282
1283         /* Not in promiscuous mode */
1284         if (priv->flags & MLX4_EN_FLAG_PROMISC)
1285                 mlx4_en_clear_promisc_mode(priv, mdev);
1286
1287         mlx4_en_do_multicast(priv, dev, mdev);
1288 out:
1289         mutex_unlock(&mdev->state_lock);
1290 }
1291
1292 #ifdef CONFIG_NET_POLL_CONTROLLER
1293 static void mlx4_en_netpoll(struct net_device *dev)
1294 {
1295         struct mlx4_en_priv *priv = netdev_priv(dev);
1296         struct mlx4_en_cq *cq;
1297         int i;
1298
1299         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1300                 cq = priv->tx_cq[TX][i];
1301                 napi_schedule(&cq->napi);
1302         }
1303 }
1304 #endif
1305
1306 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1307 {
1308         u64 reg_id;
1309         int err = 0;
1310         int *qpn = &priv->base_qpn;
1311         struct mlx4_mac_entry *entry;
1312
1313         err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1314         if (err)
1315                 return err;
1316
1317         err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1318                                        &priv->tunnel_reg_id);
1319         if (err)
1320                 goto tunnel_err;
1321
1322         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1323         if (!entry) {
1324                 err = -ENOMEM;
1325                 goto alloc_err;
1326         }
1327
1328         memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1329         memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1330         entry->reg_id = reg_id;
1331         hlist_add_head_rcu(&entry->hlist,
1332                            &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1333
1334         return 0;
1335
1336 alloc_err:
1337         if (priv->tunnel_reg_id)
1338                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1339
1340 tunnel_err:
1341         mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1342         return err;
1343 }
1344
1345 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1346 {
1347         u64 mac;
1348         unsigned int i;
1349         int qpn = priv->base_qpn;
1350         struct hlist_head *bucket;
1351         struct hlist_node *tmp;
1352         struct mlx4_mac_entry *entry;
1353
1354         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1355                 bucket = &priv->mac_hash[i];
1356                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1357                         mac = mlx4_mac_to_u64(entry->mac);
1358                         en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1359                                entry->mac);
1360                         mlx4_en_uc_steer_release(priv, entry->mac,
1361                                                  qpn, entry->reg_id);
1362
1363                         mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1364                         hlist_del_rcu(&entry->hlist);
1365                         kfree_rcu(entry, rcu);
1366                 }
1367         }
1368
1369         if (priv->tunnel_reg_id) {
1370                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1371                 priv->tunnel_reg_id = 0;
1372         }
1373 }
1374
1375 static void mlx4_en_tx_timeout(struct net_device *dev)
1376 {
1377         struct mlx4_en_priv *priv = netdev_priv(dev);
1378         struct mlx4_en_dev *mdev = priv->mdev;
1379         int i;
1380
1381         if (netif_msg_timer(priv))
1382                 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1383
1384         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1385                 struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][i];
1386
1387                 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1388                         continue;
1389                 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1390                         i, tx_ring->qpn, tx_ring->sp_cqn,
1391                         tx_ring->cons, tx_ring->prod);
1392         }
1393
1394         priv->port_stats.tx_timeout++;
1395         if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state)) {
1396                 en_dbg(DRV, priv, "Scheduling port restart\n");
1397                 queue_work(mdev->workqueue, &priv->restart_task);
1398         }
1399 }
1400
1401
1402 static void
1403 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1404 {
1405         struct mlx4_en_priv *priv = netdev_priv(dev);
1406
1407         spin_lock_bh(&priv->stats_lock);
1408         mlx4_en_fold_software_stats(dev);
1409         netdev_stats_to_stats64(stats, &dev->stats);
1410         spin_unlock_bh(&priv->stats_lock);
1411 }
1412
1413 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1414 {
1415         struct mlx4_en_cq *cq;
1416         int i, t;
1417
1418         /* If we haven't received a specific coalescing setting
1419          * (module param), we set the moderation parameters as follows:
1420          * - moder_cnt is set to the number of mtu sized packets to
1421          *   satisfy our coalescing target.
1422          * - moder_time is set to a fixed value.
1423          */
1424         priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1425         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1426         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1427         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1428         en_dbg(INTR, priv, "Default coalescing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1429                priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1430
1431         /* Setup cq moderation params */
1432         for (i = 0; i < priv->rx_ring_num; i++) {
1433                 cq = priv->rx_cq[i];
1434                 cq->moder_cnt = priv->rx_frames;
1435                 cq->moder_time = priv->rx_usecs;
1436                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1437                 priv->last_moder_packets[i] = 0;
1438                 priv->last_moder_bytes[i] = 0;
1439         }
1440
1441         for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1442                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1443                         cq = priv->tx_cq[t][i];
1444                         cq->moder_cnt = priv->tx_frames;
1445                         cq->moder_time = priv->tx_usecs;
1446                 }
1447         }
1448
1449         /* Reset auto-moderation params */
1450         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1451         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1452         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1453         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1454         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1455         priv->adaptive_rx_coal = 1;
1456         priv->last_moder_jiffies = 0;
1457         priv->last_moder_tx_packets = 0;
1458 }
1459
1460 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1461 {
1462         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1463         u32 pkt_rate_high, pkt_rate_low;
1464         struct mlx4_en_cq *cq;
1465         unsigned long packets;
1466         unsigned long rate;
1467         unsigned long avg_pkt_size;
1468         unsigned long rx_packets;
1469         unsigned long rx_bytes;
1470         unsigned long rx_pkt_diff;
1471         int moder_time;
1472         int ring, err;
1473
1474         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1475                 return;
1476
1477         pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1478         pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1479
1480         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1481                 rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1482                 rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1483
1484                 rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1485                 packets = rx_pkt_diff;
1486                 rate = packets * HZ / period;
1487                 avg_pkt_size = packets ? (rx_bytes -
1488                                 priv->last_moder_bytes[ring]) / packets : 0;
1489
1490                 /* Apply auto-moderation only when packet rate
1491                  * exceeds a rate that it matters */
1492                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1493                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1494                         if (rate <= pkt_rate_low)
1495                                 moder_time = priv->rx_usecs_low;
1496                         else if (rate >= pkt_rate_high)
1497                                 moder_time = priv->rx_usecs_high;
1498                         else
1499                                 moder_time = (rate - pkt_rate_low) *
1500                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1501                                         (pkt_rate_high - pkt_rate_low) +
1502                                         priv->rx_usecs_low;
1503                 } else {
1504                         moder_time = priv->rx_usecs_low;
1505                 }
1506
1507                 cq = priv->rx_cq[ring];
1508                 if (moder_time != priv->last_moder_time[ring] ||
1509                     cq->moder_cnt != priv->rx_frames) {
1510                         priv->last_moder_time[ring] = moder_time;
1511                         cq->moder_time = moder_time;
1512                         cq->moder_cnt = priv->rx_frames;
1513                         err = mlx4_en_set_cq_moder(priv, cq);
1514                         if (err)
1515                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1516                                        ring);
1517                 }
1518                 priv->last_moder_packets[ring] = rx_packets;
1519                 priv->last_moder_bytes[ring] = rx_bytes;
1520         }
1521
1522         priv->last_moder_jiffies = jiffies;
1523 }
1524
1525 static void mlx4_en_do_get_stats(struct work_struct *work)
1526 {
1527         struct delayed_work *delay = to_delayed_work(work);
1528         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1529                                                  stats_task);
1530         struct mlx4_en_dev *mdev = priv->mdev;
1531         int err;
1532
1533         mutex_lock(&mdev->state_lock);
1534         if (mdev->device_up) {
1535                 if (priv->port_up) {
1536                         err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1537                         if (err)
1538                                 en_dbg(HW, priv, "Could not update stats\n");
1539
1540                         mlx4_en_auto_moderation(priv);
1541                 }
1542
1543                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1544         }
1545         if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1546                 mlx4_en_do_set_mac(priv, priv->current_mac);
1547                 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1548         }
1549         mutex_unlock(&mdev->state_lock);
1550 }
1551
1552 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1553  * periodically
1554  */
1555 static void mlx4_en_service_task(struct work_struct *work)
1556 {
1557         struct delayed_work *delay = to_delayed_work(work);
1558         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1559                                                  service_task);
1560         struct mlx4_en_dev *mdev = priv->mdev;
1561
1562         mutex_lock(&mdev->state_lock);
1563         if (mdev->device_up) {
1564                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1565                         mlx4_en_ptp_overflow_check(mdev);
1566
1567                 mlx4_en_recover_from_oom(priv);
1568                 queue_delayed_work(mdev->workqueue, &priv->service_task,
1569                                    SERVICE_TASK_DELAY);
1570         }
1571         mutex_unlock(&mdev->state_lock);
1572 }
1573
1574 static void mlx4_en_linkstate(struct work_struct *work)
1575 {
1576         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1577                                                  linkstate_task);
1578         struct mlx4_en_dev *mdev = priv->mdev;
1579         int linkstate = priv->link_state;
1580
1581         mutex_lock(&mdev->state_lock);
1582         /* If observable port state changed set carrier state and
1583          * report to system log */
1584         if (priv->last_link_state != linkstate) {
1585                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1586                         en_info(priv, "Link Down\n");
1587                         netif_carrier_off(priv->dev);
1588                 } else {
1589                         en_info(priv, "Link Up\n");
1590                         netif_carrier_on(priv->dev);
1591                 }
1592         }
1593         priv->last_link_state = linkstate;
1594         mutex_unlock(&mdev->state_lock);
1595 }
1596
1597 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1598 {
1599         struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1600         int numa_node = priv->mdev->dev->numa_node;
1601
1602         if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1603                 return -ENOMEM;
1604
1605         cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1606                         ring->affinity_mask);
1607         return 0;
1608 }
1609
1610 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1611 {
1612         free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1613 }
1614
1615 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1616                                       int tx_ring_idx)
1617 {
1618         struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1619         int rr_index = tx_ring_idx;
1620
1621         tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1622         tx_ring->recycle_ring = priv->rx_ring[rr_index];
1623         en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1624                TX_XDP, tx_ring_idx, rr_index);
1625 }
1626
1627 int mlx4_en_start_port(struct net_device *dev)
1628 {
1629         struct mlx4_en_priv *priv = netdev_priv(dev);
1630         struct mlx4_en_dev *mdev = priv->mdev;
1631         struct mlx4_en_cq *cq;
1632         struct mlx4_en_tx_ring *tx_ring;
1633         int rx_index = 0;
1634         int err = 0;
1635         int i, t;
1636         int j;
1637         u8 mc_list[16] = {0};
1638
1639         if (priv->port_up) {
1640                 en_dbg(DRV, priv, "start port called while port already up\n");
1641                 return 0;
1642         }
1643
1644         INIT_LIST_HEAD(&priv->mc_list);
1645         INIT_LIST_HEAD(&priv->curr_list);
1646         INIT_LIST_HEAD(&priv->ethtool_list);
1647         memset(&priv->ethtool_rules[0], 0,
1648                sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1649
1650         /* Calculate Rx buf size */
1651         dev->mtu = min(dev->mtu, priv->max_mtu);
1652         mlx4_en_calc_rx_buf(dev);
1653         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1654
1655         /* Configure rx cq's and rings */
1656         err = mlx4_en_activate_rx_rings(priv);
1657         if (err) {
1658                 en_err(priv, "Failed to activate RX rings\n");
1659                 return err;
1660         }
1661         for (i = 0; i < priv->rx_ring_num; i++) {
1662                 cq = priv->rx_cq[i];
1663
1664                 err = mlx4_en_init_affinity_hint(priv, i);
1665                 if (err) {
1666                         en_err(priv, "Failed preparing IRQ affinity hint\n");
1667                         goto cq_err;
1668                 }
1669
1670                 err = mlx4_en_activate_cq(priv, cq, i);
1671                 if (err) {
1672                         en_err(priv, "Failed activating Rx CQ\n");
1673                         mlx4_en_free_affinity_hint(priv, i);
1674                         goto cq_err;
1675                 }
1676
1677                 for (j = 0; j < cq->size; j++) {
1678                         struct mlx4_cqe *cqe = NULL;
1679
1680                         cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1681                               priv->cqe_factor;
1682                         cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1683                 }
1684
1685                 err = mlx4_en_set_cq_moder(priv, cq);
1686                 if (err) {
1687                         en_err(priv, "Failed setting cq moderation parameters\n");
1688                         mlx4_en_deactivate_cq(priv, cq);
1689                         mlx4_en_free_affinity_hint(priv, i);
1690                         goto cq_err;
1691                 }
1692                 mlx4_en_arm_cq(priv, cq);
1693                 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1694                 ++rx_index;
1695         }
1696
1697         /* Set qp number */
1698         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1699         err = mlx4_en_get_qp(priv);
1700         if (err) {
1701                 en_err(priv, "Failed getting eth qp\n");
1702                 goto cq_err;
1703         }
1704         mdev->mac_removed[priv->port] = 0;
1705
1706         priv->counter_index =
1707                         mlx4_get_default_counter_index(mdev->dev, priv->port);
1708
1709         err = mlx4_en_config_rss_steer(priv);
1710         if (err) {
1711                 en_err(priv, "Failed configuring rss steering\n");
1712                 goto mac_err;
1713         }
1714
1715         err = mlx4_en_create_drop_qp(priv);
1716         if (err)
1717                 goto rss_err;
1718
1719         /* Configure tx cq's and rings */
1720         for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1721                 u8 num_tx_rings_p_up = t == TX ?
1722                         priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1723
1724                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1725                         /* Configure cq */
1726                         cq = priv->tx_cq[t][i];
1727                         err = mlx4_en_activate_cq(priv, cq, i);
1728                         if (err) {
1729                                 en_err(priv, "Failed allocating Tx CQ\n");
1730                                 goto tx_err;
1731                         }
1732                         err = mlx4_en_set_cq_moder(priv, cq);
1733                         if (err) {
1734                                 en_err(priv, "Failed setting cq moderation parameters\n");
1735                                 mlx4_en_deactivate_cq(priv, cq);
1736                                 goto tx_err;
1737                         }
1738                         en_dbg(DRV, priv,
1739                                "Resetting index of collapsed CQ:%d to -1\n", i);
1740                         cq->buf->wqe_index = cpu_to_be16(0xffff);
1741
1742                         /* Configure ring */
1743                         tx_ring = priv->tx_ring[t][i];
1744                         err = mlx4_en_activate_tx_ring(priv, tx_ring,
1745                                                        cq->mcq.cqn,
1746                                                        i / num_tx_rings_p_up);
1747                         if (err) {
1748                                 en_err(priv, "Failed allocating Tx ring\n");
1749                                 mlx4_en_deactivate_cq(priv, cq);
1750                                 goto tx_err;
1751                         }
1752                         clear_bit(MLX4_EN_TX_RING_STATE_RECOVERING, &tx_ring->state);
1753                         if (t != TX_XDP) {
1754                                 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1755                                 tx_ring->recycle_ring = NULL;
1756
1757                                 /* Arm CQ for TX completions */
1758                                 mlx4_en_arm_cq(priv, cq);
1759
1760                         } else {
1761                                 mlx4_en_init_recycle_ring(priv, i);
1762                                 /* XDP TX CQ should never be armed */
1763                         }
1764
1765                         /* Set initial ownership of all Tx TXBBs to SW (1) */
1766                         for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1767                                 *((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1768                 }
1769         }
1770
1771         /* Configure port */
1772         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1773                                     priv->rx_skb_size + ETH_FCS_LEN,
1774                                     priv->prof->tx_pause,
1775                                     priv->prof->tx_ppp,
1776                                     priv->prof->rx_pause,
1777                                     priv->prof->rx_ppp);
1778         if (err) {
1779                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1780                        priv->port, err);
1781                 goto tx_err;
1782         }
1783
1784         err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1785         if (err) {
1786                 en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1787                        dev->mtu, priv->port, err);
1788                 goto tx_err;
1789         }
1790
1791         /* Set default qp number */
1792         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1793         if (err) {
1794                 en_err(priv, "Failed setting default qp numbers\n");
1795                 goto tx_err;
1796         }
1797
1798         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1799                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1800                 if (err) {
1801                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1802                                err);
1803                         goto tx_err;
1804                 }
1805         }
1806
1807         /* Init port */
1808         en_dbg(HW, priv, "Initializing port\n");
1809         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1810         if (err) {
1811                 en_err(priv, "Failed Initializing port\n");
1812                 goto tx_err;
1813         }
1814
1815         /* Set Unicast and VXLAN steering rules */
1816         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1817             mlx4_en_set_rss_steer_rules(priv))
1818                 mlx4_warn(mdev, "Failed setting steering rules\n");
1819
1820         /* Attach rx QP to bradcast address */
1821         eth_broadcast_addr(&mc_list[10]);
1822         mc_list[5] = priv->port; /* needed for B0 steering support */
1823         if (mlx4_multicast_attach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1824                                   priv->port, 0, MLX4_PROT_ETH,
1825                                   &priv->broadcast_id))
1826                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1827
1828         /* Must redo promiscuous mode setup. */
1829         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1830
1831         /* Schedule multicast task to populate multicast list */
1832         queue_work(mdev->workqueue, &priv->rx_mode_task);
1833
1834         if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1835                 udp_tunnel_get_rx_info(dev);
1836
1837         priv->port_up = true;
1838
1839         /* Process all completions if exist to prevent
1840          * the queues freezing if they are full
1841          */
1842         for (i = 0; i < priv->rx_ring_num; i++) {
1843                 local_bh_disable();
1844                 napi_schedule(&priv->rx_cq[i]->napi);
1845                 local_bh_enable();
1846         }
1847
1848         clear_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state);
1849         netif_tx_start_all_queues(dev);
1850         netif_device_attach(dev);
1851
1852         return 0;
1853
1854 tx_err:
1855         if (t == MLX4_EN_NUM_TX_TYPES) {
1856                 t--;
1857                 i = priv->tx_ring_num[t];
1858         }
1859         while (t >= 0) {
1860                 while (i--) {
1861                         mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1862                         mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1863                 }
1864                 if (!t--)
1865                         break;
1866                 i = priv->tx_ring_num[t];
1867         }
1868         mlx4_en_destroy_drop_qp(priv);
1869 rss_err:
1870         mlx4_en_release_rss_steer(priv);
1871 mac_err:
1872         mlx4_en_put_qp(priv);
1873 cq_err:
1874         while (rx_index--) {
1875                 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1876                 mlx4_en_free_affinity_hint(priv, rx_index);
1877         }
1878         for (i = 0; i < priv->rx_ring_num; i++)
1879                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1880
1881         return err; /* need to close devices */
1882 }
1883
1884
1885 void mlx4_en_stop_port(struct net_device *dev, int detach)
1886 {
1887         struct mlx4_en_priv *priv = netdev_priv(dev);
1888         struct mlx4_en_dev *mdev = priv->mdev;
1889         struct mlx4_en_mc_list *mclist, *tmp;
1890         struct ethtool_flow_id *flow, *tmp_flow;
1891         int i, t;
1892         u8 mc_list[16] = {0};
1893
1894         if (!priv->port_up) {
1895                 en_dbg(DRV, priv, "stop port called while port already down\n");
1896                 return;
1897         }
1898
1899         /* close port*/
1900         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1901
1902         /* Synchronize with tx routine */
1903         netif_tx_lock_bh(dev);
1904         if (detach)
1905                 netif_device_detach(dev);
1906         netif_tx_stop_all_queues(dev);
1907         netif_tx_unlock_bh(dev);
1908
1909         netif_tx_disable(dev);
1910
1911         spin_lock_bh(&priv->stats_lock);
1912         mlx4_en_fold_software_stats(dev);
1913         /* Set port as not active */
1914         priv->port_up = false;
1915         spin_unlock_bh(&priv->stats_lock);
1916
1917         priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1918
1919         /* Promsicuous mode */
1920         if (mdev->dev->caps.steering_mode ==
1921             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1922                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1923                                  MLX4_EN_FLAG_MC_PROMISC);
1924                 mlx4_flow_steer_promisc_remove(mdev->dev,
1925                                                priv->port,
1926                                                MLX4_FS_ALL_DEFAULT);
1927                 mlx4_flow_steer_promisc_remove(mdev->dev,
1928                                                priv->port,
1929                                                MLX4_FS_MC_DEFAULT);
1930         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1931                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1932
1933                 /* Disable promiscouos mode */
1934                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1935                                             priv->port);
1936
1937                 /* Disable Multicast promisc */
1938                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1939                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1940                                                       priv->port);
1941                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1942                 }
1943         }
1944
1945         /* Detach All multicasts */
1946         eth_broadcast_addr(&mc_list[10]);
1947         mc_list[5] = priv->port; /* needed for B0 steering support */
1948         mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1949                               MLX4_PROT_ETH, priv->broadcast_id);
1950         list_for_each_entry(mclist, &priv->curr_list, list) {
1951                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1952                 mc_list[5] = priv->port;
1953                 mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp,
1954                                       mc_list, MLX4_PROT_ETH, mclist->reg_id);
1955                 if (mclist->tunnel_reg_id)
1956                         mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1957         }
1958         mlx4_en_clear_list(dev);
1959         list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1960                 list_del(&mclist->list);
1961                 kfree(mclist);
1962         }
1963
1964         /* Flush multicast filter */
1965         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1966
1967         /* Remove flow steering rules for the port*/
1968         if (mdev->dev->caps.steering_mode ==
1969             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1970                 ASSERT_RTNL();
1971                 list_for_each_entry_safe(flow, tmp_flow,
1972                                          &priv->ethtool_list, list) {
1973                         mlx4_flow_detach(mdev->dev, flow->id);
1974                         list_del(&flow->list);
1975                 }
1976         }
1977
1978         mlx4_en_destroy_drop_qp(priv);
1979
1980         /* Free TX Rings */
1981         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1982                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1983                         mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1984                         mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1985                 }
1986         }
1987         msleep(10);
1988
1989         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1990                 for (i = 0; i < priv->tx_ring_num[t]; i++)
1991                         mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1992
1993         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1994                 mlx4_en_delete_rss_steer_rules(priv);
1995
1996         /* Free RSS qps */
1997         mlx4_en_release_rss_steer(priv);
1998
1999         /* Unregister Mac address for the port */
2000         mlx4_en_put_qp(priv);
2001         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
2002                 mdev->mac_removed[priv->port] = 1;
2003
2004         /* Free RX Rings */
2005         for (i = 0; i < priv->rx_ring_num; i++) {
2006                 struct mlx4_en_cq *cq = priv->rx_cq[i];
2007
2008                 napi_synchronize(&cq->napi);
2009                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
2010                 mlx4_en_deactivate_cq(priv, cq);
2011
2012                 mlx4_en_free_affinity_hint(priv, i);
2013         }
2014 }
2015
2016 static void mlx4_en_restart(struct work_struct *work)
2017 {
2018         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2019                                                  restart_task);
2020         struct mlx4_en_dev *mdev = priv->mdev;
2021         struct net_device *dev = priv->dev;
2022
2023         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
2024
2025         rtnl_lock();
2026         mutex_lock(&mdev->state_lock);
2027         if (priv->port_up) {
2028                 mlx4_en_stop_port(dev, 1);
2029                 if (mlx4_en_start_port(dev))
2030                         en_err(priv, "Failed restarting port %d\n", priv->port);
2031         }
2032         mutex_unlock(&mdev->state_lock);
2033         rtnl_unlock();
2034 }
2035
2036 static void mlx4_en_clear_stats(struct net_device *dev)
2037 {
2038         struct mlx4_en_priv *priv = netdev_priv(dev);
2039         struct mlx4_en_dev *mdev = priv->mdev;
2040         struct mlx4_en_tx_ring **tx_ring;
2041         int i;
2042
2043         if (!mlx4_is_slave(mdev->dev))
2044                 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
2045                         en_dbg(HW, priv, "Failed dumping statistics\n");
2046
2047         memset(&priv->pstats, 0, sizeof(priv->pstats));
2048         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
2049         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
2050         memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
2051         memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
2052         memset(&priv->rx_priority_flowstats, 0,
2053                sizeof(priv->rx_priority_flowstats));
2054         memset(&priv->tx_priority_flowstats, 0,
2055                sizeof(priv->tx_priority_flowstats));
2056         memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
2057
2058         tx_ring = priv->tx_ring[TX];
2059         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
2060                 tx_ring[i]->bytes = 0;
2061                 tx_ring[i]->packets = 0;
2062                 tx_ring[i]->tx_csum = 0;
2063                 tx_ring[i]->tx_dropped = 0;
2064                 tx_ring[i]->queue_stopped = 0;
2065                 tx_ring[i]->wake_queue = 0;
2066                 tx_ring[i]->tso_packets = 0;
2067                 tx_ring[i]->xmit_more = 0;
2068         }
2069         for (i = 0; i < priv->rx_ring_num; i++) {
2070                 priv->rx_ring[i]->bytes = 0;
2071                 priv->rx_ring[i]->packets = 0;
2072                 priv->rx_ring[i]->csum_ok = 0;
2073                 priv->rx_ring[i]->csum_none = 0;
2074                 priv->rx_ring[i]->csum_complete = 0;
2075         }
2076 }
2077
2078 static int mlx4_en_open(struct net_device *dev)
2079 {
2080         struct mlx4_en_priv *priv = netdev_priv(dev);
2081         struct mlx4_en_dev *mdev = priv->mdev;
2082         int err = 0;
2083
2084         mutex_lock(&mdev->state_lock);
2085
2086         if (!mdev->device_up) {
2087                 en_err(priv, "Cannot open - device down/disabled\n");
2088                 err = -EBUSY;
2089                 goto out;
2090         }
2091
2092         /* Reset HW statistics and SW counters */
2093         mlx4_en_clear_stats(dev);
2094
2095         err = mlx4_en_start_port(dev);
2096         if (err)
2097                 en_err(priv, "Failed starting port:%d\n", priv->port);
2098
2099 out:
2100         mutex_unlock(&mdev->state_lock);
2101         return err;
2102 }
2103
2104
2105 static int mlx4_en_close(struct net_device *dev)
2106 {
2107         struct mlx4_en_priv *priv = netdev_priv(dev);
2108         struct mlx4_en_dev *mdev = priv->mdev;
2109
2110         en_dbg(IFDOWN, priv, "Close port called\n");
2111
2112         mutex_lock(&mdev->state_lock);
2113
2114         mlx4_en_stop_port(dev, 0);
2115         netif_carrier_off(dev);
2116
2117         mutex_unlock(&mdev->state_lock);
2118         return 0;
2119 }
2120
2121 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2122 {
2123         int i, t;
2124
2125 #ifdef CONFIG_RFS_ACCEL
2126         priv->dev->rx_cpu_rmap = NULL;
2127 #endif
2128
2129         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2130                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2131                         if (priv->tx_ring[t] && priv->tx_ring[t][i])
2132                                 mlx4_en_destroy_tx_ring(priv,
2133                                                         &priv->tx_ring[t][i]);
2134                         if (priv->tx_cq[t] && priv->tx_cq[t][i])
2135                                 mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2136                 }
2137                 kfree(priv->tx_ring[t]);
2138                 kfree(priv->tx_cq[t]);
2139         }
2140
2141         for (i = 0; i < priv->rx_ring_num; i++) {
2142                 if (priv->rx_ring[i])
2143                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2144                                 priv->prof->rx_ring_size, priv->stride);
2145                 if (priv->rx_cq[i])
2146                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2147         }
2148
2149 }
2150
2151 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2152 {
2153         struct mlx4_en_port_profile *prof = priv->prof;
2154         int i, t;
2155         int node;
2156
2157         /* Create tx Rings */
2158         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2159                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2160                         node = cpu_to_node(i % num_online_cpus());
2161                         if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2162                                               prof->tx_ring_size, i, t, node))
2163                                 goto err;
2164
2165                         if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2166                                                    prof->tx_ring_size,
2167                                                    TXBB_SIZE, node, i))
2168                                 goto err;
2169                 }
2170         }
2171
2172         /* Create rx Rings */
2173         for (i = 0; i < priv->rx_ring_num; i++) {
2174                 node = cpu_to_node(i % num_online_cpus());
2175                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2176                                       prof->rx_ring_size, i, RX, node))
2177                         goto err;
2178
2179                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2180                                            prof->rx_ring_size, priv->stride,
2181                                            node))
2182                         goto err;
2183         }
2184
2185 #ifdef CONFIG_RFS_ACCEL
2186         priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2187 #endif
2188
2189         return 0;
2190
2191 err:
2192         en_err(priv, "Failed to allocate NIC resources\n");
2193         for (i = 0; i < priv->rx_ring_num; i++) {
2194                 if (priv->rx_ring[i])
2195                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2196                                                 prof->rx_ring_size,
2197                                                 priv->stride);
2198                 if (priv->rx_cq[i])
2199                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2200         }
2201         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2202                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2203                         if (priv->tx_ring[t][i])
2204                                 mlx4_en_destroy_tx_ring(priv,
2205                                                         &priv->tx_ring[t][i]);
2206                         if (priv->tx_cq[t][i])
2207                                 mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2208                 }
2209         }
2210         return -ENOMEM;
2211 }
2212
2213
2214 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2215                              struct mlx4_en_priv *src,
2216                              struct mlx4_en_port_profile *prof)
2217 {
2218         int t;
2219
2220         memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2221                sizeof(dst->hwtstamp_config));
2222         dst->num_tx_rings_p_up = prof->num_tx_rings_p_up;
2223         dst->rx_ring_num = prof->rx_ring_num;
2224         dst->flags = prof->flags;
2225         dst->mdev = src->mdev;
2226         dst->port = src->port;
2227         dst->dev = src->dev;
2228         dst->prof = prof;
2229         dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2230                                          DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2231
2232         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2233                 dst->tx_ring_num[t] = prof->tx_ring_num[t];
2234                 if (!dst->tx_ring_num[t])
2235                         continue;
2236
2237                 dst->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
2238                                           MAX_TX_RINGS, GFP_KERNEL);
2239                 if (!dst->tx_ring[t])
2240                         goto err_free_tx;
2241
2242                 dst->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
2243                                         MAX_TX_RINGS, GFP_KERNEL);
2244                 if (!dst->tx_cq[t]) {
2245                         kfree(dst->tx_ring[t]);
2246                         goto err_free_tx;
2247                 }
2248         }
2249
2250         return 0;
2251
2252 err_free_tx:
2253         while (t--) {
2254                 kfree(dst->tx_ring[t]);
2255                 kfree(dst->tx_cq[t]);
2256         }
2257         return -ENOMEM;
2258 }
2259
2260 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2261                                 struct mlx4_en_priv *src)
2262 {
2263         int t;
2264         memcpy(dst->rx_ring, src->rx_ring,
2265                sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2266         memcpy(dst->rx_cq, src->rx_cq,
2267                sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2268         memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2269                sizeof(dst->hwtstamp_config));
2270         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2271                 dst->tx_ring_num[t] = src->tx_ring_num[t];
2272                 dst->tx_ring[t] = src->tx_ring[t];
2273                 dst->tx_cq[t] = src->tx_cq[t];
2274         }
2275         dst->num_tx_rings_p_up = src->num_tx_rings_p_up;
2276         dst->rx_ring_num = src->rx_ring_num;
2277         memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2278 }
2279
2280 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2281                                 struct mlx4_en_priv *tmp,
2282                                 struct mlx4_en_port_profile *prof,
2283                                 bool carry_xdp_prog)
2284 {
2285         struct bpf_prog *xdp_prog;
2286         int i, t, ret;
2287
2288         ret = mlx4_en_copy_priv(tmp, priv, prof);
2289         if (ret) {
2290                 en_warn(priv, "%s: mlx4_en_copy_priv() failed, return\n",
2291                         __func__);
2292                 return ret;
2293         }
2294
2295         if (mlx4_en_alloc_resources(tmp)) {
2296                 en_warn(priv,
2297                         "%s: Resource allocation failed, using previous configuration\n",
2298                         __func__);
2299                 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2300                         kfree(tmp->tx_ring[t]);
2301                         kfree(tmp->tx_cq[t]);
2302                 }
2303                 return -ENOMEM;
2304         }
2305
2306         /* All rx_rings has the same xdp_prog.  Pick the first one. */
2307         xdp_prog = rcu_dereference_protected(
2308                 priv->rx_ring[0]->xdp_prog,
2309                 lockdep_is_held(&priv->mdev->state_lock));
2310
2311         if (xdp_prog && carry_xdp_prog) {
2312                 xdp_prog = bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2313                 if (IS_ERR(xdp_prog)) {
2314                         mlx4_en_free_resources(tmp);
2315                         return PTR_ERR(xdp_prog);
2316                 }
2317                 for (i = 0; i < tmp->rx_ring_num; i++)
2318                         rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2319                                            xdp_prog);
2320         }
2321
2322         return 0;
2323 }
2324
2325 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2326                                     struct mlx4_en_priv *tmp)
2327 {
2328         mlx4_en_free_resources(priv);
2329         mlx4_en_update_priv(priv, tmp);
2330 }
2331
2332 void mlx4_en_destroy_netdev(struct net_device *dev)
2333 {
2334         struct mlx4_en_priv *priv = netdev_priv(dev);
2335         struct mlx4_en_dev *mdev = priv->mdev;
2336
2337         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2338
2339         /* Unregister device - this will close the port if it was up */
2340         if (priv->registered) {
2341                 devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2342                                                               priv->port));
2343                 unregister_netdev(dev);
2344         }
2345
2346         if (priv->allocated)
2347                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2348
2349         cancel_delayed_work(&priv->stats_task);
2350         cancel_delayed_work(&priv->service_task);
2351         /* flush any pending task for this netdev */
2352         flush_workqueue(mdev->workqueue);
2353
2354         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2355                 mlx4_en_remove_timestamp(mdev);
2356
2357         /* Detach the netdev so tasks would not attempt to access it */
2358         mutex_lock(&mdev->state_lock);
2359         mdev->pndev[priv->port] = NULL;
2360         mdev->upper[priv->port] = NULL;
2361
2362 #ifdef CONFIG_RFS_ACCEL
2363         mlx4_en_cleanup_filters(priv);
2364 #endif
2365
2366         mlx4_en_free_resources(priv);
2367         mutex_unlock(&mdev->state_lock);
2368
2369         free_netdev(dev);
2370 }
2371
2372 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2373 {
2374         struct mlx4_en_priv *priv = netdev_priv(dev);
2375
2376         if (mtu > MLX4_EN_MAX_XDP_MTU) {
2377                 en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2378                        mtu, MLX4_EN_MAX_XDP_MTU);
2379                 return false;
2380         }
2381
2382         return true;
2383 }
2384
2385 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2386 {
2387         struct mlx4_en_priv *priv = netdev_priv(dev);
2388         struct mlx4_en_dev *mdev = priv->mdev;
2389         int err = 0;
2390
2391         en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2392                  dev->mtu, new_mtu);
2393
2394         if (priv->tx_ring_num[TX_XDP] &&
2395             !mlx4_en_check_xdp_mtu(dev, new_mtu))
2396                 return -EOPNOTSUPP;
2397
2398         dev->mtu = new_mtu;
2399
2400         if (netif_running(dev)) {
2401                 mutex_lock(&mdev->state_lock);
2402                 if (!mdev->device_up) {
2403                         /* NIC is probably restarting - let restart task reset
2404                          * the port */
2405                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2406                 } else {
2407                         mlx4_en_stop_port(dev, 1);
2408                         err = mlx4_en_start_port(dev);
2409                         if (err) {
2410                                 en_err(priv, "Failed restarting port:%d\n",
2411                                          priv->port);
2412                                 if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING,
2413                                                       &priv->state))
2414                                         queue_work(mdev->workqueue, &priv->restart_task);
2415                         }
2416                 }
2417                 mutex_unlock(&mdev->state_lock);
2418         }
2419         return 0;
2420 }
2421
2422 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2423 {
2424         struct mlx4_en_priv *priv = netdev_priv(dev);
2425         struct mlx4_en_dev *mdev = priv->mdev;
2426         struct hwtstamp_config config;
2427
2428         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2429                 return -EFAULT;
2430
2431         /* reserved for future extensions */
2432         if (config.flags)
2433                 return -EINVAL;
2434
2435         /* device doesn't support time stamping */
2436         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2437                 return -EINVAL;
2438
2439         /* TX HW timestamp */
2440         switch (config.tx_type) {
2441         case HWTSTAMP_TX_OFF:
2442         case HWTSTAMP_TX_ON:
2443                 break;
2444         default:
2445                 return -ERANGE;
2446         }
2447
2448         /* RX HW timestamp */
2449         switch (config.rx_filter) {
2450         case HWTSTAMP_FILTER_NONE:
2451                 break;
2452         case HWTSTAMP_FILTER_ALL:
2453         case HWTSTAMP_FILTER_SOME:
2454         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2455         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2456         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2457         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2458         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2459         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2460         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2461         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2462         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2463         case HWTSTAMP_FILTER_PTP_V2_EVENT:
2464         case HWTSTAMP_FILTER_PTP_V2_SYNC:
2465         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2466         case HWTSTAMP_FILTER_NTP_ALL:
2467                 config.rx_filter = HWTSTAMP_FILTER_ALL;
2468                 break;
2469         default:
2470                 return -ERANGE;
2471         }
2472
2473         if (mlx4_en_reset_config(dev, config, dev->features)) {
2474                 config.tx_type = HWTSTAMP_TX_OFF;
2475                 config.rx_filter = HWTSTAMP_FILTER_NONE;
2476         }
2477
2478         return copy_to_user(ifr->ifr_data, &config,
2479                             sizeof(config)) ? -EFAULT : 0;
2480 }
2481
2482 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2483 {
2484         struct mlx4_en_priv *priv = netdev_priv(dev);
2485
2486         return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2487                             sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2488 }
2489
2490 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2491 {
2492         switch (cmd) {
2493         case SIOCSHWTSTAMP:
2494                 return mlx4_en_hwtstamp_set(dev, ifr);
2495         case SIOCGHWTSTAMP:
2496                 return mlx4_en_hwtstamp_get(dev, ifr);
2497         default:
2498                 return -EOPNOTSUPP;
2499         }
2500 }
2501
2502 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2503                                               netdev_features_t features)
2504 {
2505         struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2506         struct mlx4_en_dev *mdev = en_priv->mdev;
2507
2508         /* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2509          * enable/disable make sure S-TAG flag is always in same state as
2510          * C-TAG.
2511          */
2512         if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2513             !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2514                 features |= NETIF_F_HW_VLAN_STAG_RX;
2515         else
2516                 features &= ~NETIF_F_HW_VLAN_STAG_RX;
2517
2518         return features;
2519 }
2520
2521 static int mlx4_en_set_features(struct net_device *netdev,
2522                 netdev_features_t features)
2523 {
2524         struct mlx4_en_priv *priv = netdev_priv(netdev);
2525         bool reset = false;
2526         int ret = 0;
2527
2528         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2529                 en_info(priv, "Turn %s RX-FCS\n",
2530                         (features & NETIF_F_RXFCS) ? "ON" : "OFF");
2531                 reset = true;
2532         }
2533
2534         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2535                 u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2536
2537                 en_info(priv, "Turn %s RX-ALL\n",
2538                         ignore_fcs_value ? "ON" : "OFF");
2539                 ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2540                                               priv->port, ignore_fcs_value);
2541                 if (ret)
2542                         return ret;
2543         }
2544
2545         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2546                 en_info(priv, "Turn %s RX vlan strip offload\n",
2547                         (features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2548                 reset = true;
2549         }
2550
2551         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2552                 en_info(priv, "Turn %s TX vlan strip offload\n",
2553                         (features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2554
2555         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2556                 en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2557                         (features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2558
2559         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2560                 en_info(priv, "Turn %s loopback\n",
2561                         (features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2562                 mlx4_en_update_loopback_state(netdev, features);
2563         }
2564
2565         if (reset) {
2566                 ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2567                                            features);
2568                 if (ret)
2569                         return ret;
2570         }
2571
2572         return 0;
2573 }
2574
2575 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2576 {
2577         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2578         struct mlx4_en_dev *mdev = en_priv->mdev;
2579
2580         return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac);
2581 }
2582
2583 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2584                                __be16 vlan_proto)
2585 {
2586         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2587         struct mlx4_en_dev *mdev = en_priv->mdev;
2588
2589         return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2590                                 vlan_proto);
2591 }
2592
2593 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2594                                int max_tx_rate)
2595 {
2596         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2597         struct mlx4_en_dev *mdev = en_priv->mdev;
2598
2599         return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2600                                 max_tx_rate);
2601 }
2602
2603 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2604 {
2605         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2606         struct mlx4_en_dev *mdev = en_priv->mdev;
2607
2608         return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2609 }
2610
2611 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2612 {
2613         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2614         struct mlx4_en_dev *mdev = en_priv->mdev;
2615
2616         return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2617 }
2618
2619 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2620 {
2621         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2622         struct mlx4_en_dev *mdev = en_priv->mdev;
2623
2624         return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2625 }
2626
2627 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2628                                 struct ifla_vf_stats *vf_stats)
2629 {
2630         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2631         struct mlx4_en_dev *mdev = en_priv->mdev;
2632
2633         return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2634 }
2635
2636 #define PORT_ID_BYTE_LEN 8
2637 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2638                                     struct netdev_phys_item_id *ppid)
2639 {
2640         struct mlx4_en_priv *priv = netdev_priv(dev);
2641         struct mlx4_dev *mdev = priv->mdev->dev;
2642         int i;
2643         u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2644
2645         if (!phys_port_id)
2646                 return -EOPNOTSUPP;
2647
2648         ppid->id_len = sizeof(phys_port_id);
2649         for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2650                 ppid->id[i] =  phys_port_id & 0xff;
2651                 phys_port_id >>= 8;
2652         }
2653         return 0;
2654 }
2655
2656 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2657 {
2658         int ret;
2659         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2660                                                  vxlan_add_task);
2661
2662         ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2663         if (ret)
2664                 goto out;
2665
2666         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2667                                   VXLAN_STEER_BY_OUTER_MAC, 1);
2668 out:
2669         if (ret) {
2670                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2671                 return;
2672         }
2673
2674         /* set offloads */
2675         priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2676                                       NETIF_F_RXCSUM |
2677                                       NETIF_F_TSO | NETIF_F_TSO6 |
2678                                       NETIF_F_GSO_UDP_TUNNEL |
2679                                       NETIF_F_GSO_UDP_TUNNEL_CSUM |
2680                                       NETIF_F_GSO_PARTIAL;
2681 }
2682
2683 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2684 {
2685         int ret;
2686         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2687                                                  vxlan_del_task);
2688         /* unset offloads */
2689         priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2690                                         NETIF_F_RXCSUM |
2691                                         NETIF_F_TSO | NETIF_F_TSO6 |
2692                                         NETIF_F_GSO_UDP_TUNNEL |
2693                                         NETIF_F_GSO_UDP_TUNNEL_CSUM |
2694                                         NETIF_F_GSO_PARTIAL);
2695
2696         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2697                                   VXLAN_STEER_BY_OUTER_MAC, 0);
2698         if (ret)
2699                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2700
2701         priv->vxlan_port = 0;
2702 }
2703
2704 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2705                                    struct udp_tunnel_info *ti)
2706 {
2707         struct mlx4_en_priv *priv = netdev_priv(dev);
2708         __be16 port = ti->port;
2709         __be16 current_port;
2710
2711         if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2712                 return;
2713
2714         if (ti->sa_family != AF_INET)
2715                 return;
2716
2717         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2718                 return;
2719
2720         current_port = priv->vxlan_port;
2721         if (current_port && current_port != port) {
2722                 en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2723                         ntohs(current_port), ntohs(port));
2724                 return;
2725         }
2726
2727         priv->vxlan_port = port;
2728         queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2729 }
2730
2731 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2732                                    struct udp_tunnel_info *ti)
2733 {
2734         struct mlx4_en_priv *priv = netdev_priv(dev);
2735         __be16 port = ti->port;
2736         __be16 current_port;
2737
2738         if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2739                 return;
2740
2741         if (ti->sa_family != AF_INET)
2742                 return;
2743
2744         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2745                 return;
2746
2747         current_port = priv->vxlan_port;
2748         if (current_port != port) {
2749                 en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2750                 return;
2751         }
2752
2753         queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2754 }
2755
2756 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2757                                                 struct net_device *dev,
2758                                                 netdev_features_t features)
2759 {
2760         features = vlan_features_check(skb, features);
2761         features = vxlan_features_check(skb, features);
2762
2763         /* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2764          * support inner IPv6 checksums and segmentation so  we need to
2765          * strip that feature if this is an IPv6 encapsulated frame.
2766          */
2767         if (skb->encapsulation &&
2768             (skb->ip_summed == CHECKSUM_PARTIAL)) {
2769                 struct mlx4_en_priv *priv = netdev_priv(dev);
2770
2771                 if (!priv->vxlan_port ||
2772                     (ip_hdr(skb)->version != 4) ||
2773                     (udp_hdr(skb)->dest != priv->vxlan_port))
2774                         features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2775         }
2776
2777         return features;
2778 }
2779
2780 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2781 {
2782         struct mlx4_en_priv *priv = netdev_priv(dev);
2783         struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2784         struct mlx4_update_qp_params params;
2785         int err;
2786
2787         if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2788                 return -EOPNOTSUPP;
2789
2790         /* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2791         if (maxrate >> 12) {
2792                 params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2793                 params.rate_val  = maxrate / 1000;
2794         } else if (maxrate) {
2795                 params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2796                 params.rate_val  = maxrate;
2797         } else { /* zero serves to revoke the QP rate-limitation */
2798                 params.rate_unit = 0;
2799                 params.rate_val  = 0;
2800         }
2801
2802         err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2803                              &params);
2804         return err;
2805 }
2806
2807 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2808 {
2809         struct mlx4_en_priv *priv = netdev_priv(dev);
2810         struct mlx4_en_dev *mdev = priv->mdev;
2811         struct mlx4_en_port_profile new_prof;
2812         struct bpf_prog *old_prog;
2813         struct mlx4_en_priv *tmp;
2814         int tx_changed = 0;
2815         int xdp_ring_num;
2816         int port_up = 0;
2817         int err;
2818         int i;
2819
2820         xdp_ring_num = prog ? priv->rx_ring_num : 0;
2821
2822         /* No need to reconfigure buffers when simply swapping the
2823          * program for a new one.
2824          */
2825         if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2826                 if (prog) {
2827                         prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2828                         if (IS_ERR(prog))
2829                                 return PTR_ERR(prog);
2830                 }
2831                 mutex_lock(&mdev->state_lock);
2832                 for (i = 0; i < priv->rx_ring_num; i++) {
2833                         old_prog = rcu_dereference_protected(
2834                                         priv->rx_ring[i]->xdp_prog,
2835                                         lockdep_is_held(&mdev->state_lock));
2836                         rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2837                         if (old_prog)
2838                                 bpf_prog_put(old_prog);
2839                 }
2840                 mutex_unlock(&mdev->state_lock);
2841                 return 0;
2842         }
2843
2844         if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2845                 return -EOPNOTSUPP;
2846
2847         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2848         if (!tmp)
2849                 return -ENOMEM;
2850
2851         if (prog) {
2852                 prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2853                 if (IS_ERR(prog)) {
2854                         err = PTR_ERR(prog);
2855                         goto out;
2856                 }
2857         }
2858
2859         mutex_lock(&mdev->state_lock);
2860         memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2861         new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2862
2863         if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2864                 tx_changed = 1;
2865                 new_prof.tx_ring_num[TX] =
2866                         MAX_TX_RINGS - ALIGN(xdp_ring_num, priv->prof->num_up);
2867                 en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2868         }
2869
2870         err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2871         if (err) {
2872                 if (prog)
2873                         bpf_prog_sub(prog, priv->rx_ring_num - 1);
2874                 goto unlock_out;
2875         }
2876
2877         if (priv->port_up) {
2878                 port_up = 1;
2879                 mlx4_en_stop_port(dev, 1);
2880         }
2881
2882         mlx4_en_safe_replace_resources(priv, tmp);
2883         if (tx_changed)
2884                 netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2885
2886         for (i = 0; i < priv->rx_ring_num; i++) {
2887                 old_prog = rcu_dereference_protected(
2888                                         priv->rx_ring[i]->xdp_prog,
2889                                         lockdep_is_held(&mdev->state_lock));
2890                 rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2891                 if (old_prog)
2892                         bpf_prog_put(old_prog);
2893         }
2894
2895         if (port_up) {
2896                 err = mlx4_en_start_port(dev);
2897                 if (err) {
2898                         en_err(priv, "Failed starting port %d for XDP change\n",
2899                                priv->port);
2900                         if (!test_and_set_bit(MLX4_EN_STATE_FLAG_RESTARTING, &priv->state))
2901                                 queue_work(mdev->workqueue, &priv->restart_task);
2902                 }
2903         }
2904
2905 unlock_out:
2906         mutex_unlock(&mdev->state_lock);
2907 out:
2908         kfree(tmp);
2909         return err;
2910 }
2911
2912 static u32 mlx4_xdp_query(struct net_device *dev)
2913 {
2914         struct mlx4_en_priv *priv = netdev_priv(dev);
2915         struct mlx4_en_dev *mdev = priv->mdev;
2916         const struct bpf_prog *xdp_prog;
2917         u32 prog_id = 0;
2918
2919         if (!priv->tx_ring_num[TX_XDP])
2920                 return prog_id;
2921
2922         mutex_lock(&mdev->state_lock);
2923         xdp_prog = rcu_dereference_protected(
2924                 priv->rx_ring[0]->xdp_prog,
2925                 lockdep_is_held(&mdev->state_lock));
2926         if (xdp_prog)
2927                 prog_id = xdp_prog->aux->id;
2928         mutex_unlock(&mdev->state_lock);
2929
2930         return prog_id;
2931 }
2932
2933 static int mlx4_xdp(struct net_device *dev, struct netdev_xdp *xdp)
2934 {
2935         switch (xdp->command) {
2936         case XDP_SETUP_PROG:
2937                 return mlx4_xdp_set(dev, xdp->prog);
2938         case XDP_QUERY_PROG:
2939                 xdp->prog_id = mlx4_xdp_query(dev);
2940                 xdp->prog_attached = !!xdp->prog_id;
2941                 return 0;
2942         default:
2943                 return -EINVAL;
2944         }
2945 }
2946
2947 static const struct net_device_ops mlx4_netdev_ops = {
2948         .ndo_open               = mlx4_en_open,
2949         .ndo_stop               = mlx4_en_close,
2950         .ndo_start_xmit         = mlx4_en_xmit,
2951         .ndo_select_queue       = mlx4_en_select_queue,
2952         .ndo_get_stats64        = mlx4_en_get_stats64,
2953         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2954         .ndo_set_mac_address    = mlx4_en_set_mac,
2955         .ndo_validate_addr      = eth_validate_addr,
2956         .ndo_change_mtu         = mlx4_en_change_mtu,
2957         .ndo_do_ioctl           = mlx4_en_ioctl,
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 #ifdef CONFIG_NET_POLL_CONTROLLER
2962         .ndo_poll_controller    = mlx4_en_netpoll,
2963 #endif
2964         .ndo_set_features       = mlx4_en_set_features,
2965         .ndo_fix_features       = mlx4_en_fix_features,
2966         .ndo_setup_tc           = __mlx4_en_setup_tc,
2967 #ifdef CONFIG_RFS_ACCEL
2968         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2969 #endif
2970         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2971         .ndo_udp_tunnel_add     = mlx4_en_add_vxlan_port,
2972         .ndo_udp_tunnel_del     = mlx4_en_del_vxlan_port,
2973         .ndo_features_check     = mlx4_en_features_check,
2974         .ndo_set_tx_maxrate     = mlx4_en_set_tx_maxrate,
2975         .ndo_xdp                = mlx4_xdp,
2976 };
2977
2978 static const struct net_device_ops mlx4_netdev_ops_master = {
2979         .ndo_open               = mlx4_en_open,
2980         .ndo_stop               = mlx4_en_close,
2981         .ndo_start_xmit         = mlx4_en_xmit,
2982         .ndo_select_queue       = mlx4_en_select_queue,
2983         .ndo_get_stats64        = mlx4_en_get_stats64,
2984         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2985         .ndo_set_mac_address    = mlx4_en_set_mac,
2986         .ndo_validate_addr      = eth_validate_addr,
2987         .ndo_change_mtu         = mlx4_en_change_mtu,
2988         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2989         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2990         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2991         .ndo_set_vf_mac         = mlx4_en_set_vf_mac,
2992         .ndo_set_vf_vlan        = mlx4_en_set_vf_vlan,
2993         .ndo_set_vf_rate        = mlx4_en_set_vf_rate,
2994         .ndo_set_vf_spoofchk    = mlx4_en_set_vf_spoofchk,
2995         .ndo_set_vf_link_state  = mlx4_en_set_vf_link_state,
2996         .ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2997         .ndo_get_vf_config      = mlx4_en_get_vf_config,
2998 #ifdef CONFIG_NET_POLL_CONTROLLER
2999         .ndo_poll_controller    = mlx4_en_netpoll,
3000 #endif
3001         .ndo_set_features       = mlx4_en_set_features,
3002         .ndo_fix_features       = mlx4_en_fix_features,
3003         .ndo_setup_tc           = __mlx4_en_setup_tc,
3004 #ifdef CONFIG_RFS_ACCEL
3005         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
3006 #endif
3007         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
3008         .ndo_udp_tunnel_add     = mlx4_en_add_vxlan_port,
3009         .ndo_udp_tunnel_del     = mlx4_en_del_vxlan_port,
3010         .ndo_features_check     = mlx4_en_features_check,
3011         .ndo_set_tx_maxrate     = mlx4_en_set_tx_maxrate,
3012         .ndo_xdp                = mlx4_xdp,
3013 };
3014
3015 struct mlx4_en_bond {
3016         struct work_struct work;
3017         struct mlx4_en_priv *priv;
3018         int is_bonded;
3019         struct mlx4_port_map port_map;
3020 };
3021
3022 static void mlx4_en_bond_work(struct work_struct *work)
3023 {
3024         struct mlx4_en_bond *bond = container_of(work,
3025                                                      struct mlx4_en_bond,
3026                                                      work);
3027         int err = 0;
3028         struct mlx4_dev *dev = bond->priv->mdev->dev;
3029
3030         if (bond->is_bonded) {
3031                 if (!mlx4_is_bonded(dev)) {
3032                         err = mlx4_bond(dev);
3033                         if (err)
3034                                 en_err(bond->priv, "Fail to bond device\n");
3035                 }
3036                 if (!err) {
3037                         err = mlx4_port_map_set(dev, &bond->port_map);
3038                         if (err)
3039                                 en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
3040                                        bond->port_map.port1,
3041                                        bond->port_map.port2,
3042                                        err);
3043                 }
3044         } else if (mlx4_is_bonded(dev)) {
3045                 err = mlx4_unbond(dev);
3046                 if (err)
3047                         en_err(bond->priv, "Fail to unbond device\n");
3048         }
3049         dev_put(bond->priv->dev);
3050         kfree(bond);
3051 }
3052
3053 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
3054                                    u8 v2p_p1, u8 v2p_p2)
3055 {
3056         struct mlx4_en_bond *bond = NULL;
3057
3058         bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
3059         if (!bond)
3060                 return -ENOMEM;
3061
3062         INIT_WORK(&bond->work, mlx4_en_bond_work);
3063         bond->priv = priv;
3064         bond->is_bonded = is_bonded;
3065         bond->port_map.port1 = v2p_p1;
3066         bond->port_map.port2 = v2p_p2;
3067         dev_hold(priv->dev);
3068         queue_work(priv->mdev->workqueue, &bond->work);
3069         return 0;
3070 }
3071
3072 int mlx4_en_netdev_event(struct notifier_block *this,
3073                          unsigned long event, void *ptr)
3074 {
3075         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3076         u8 port = 0;
3077         struct mlx4_en_dev *mdev;
3078         struct mlx4_dev *dev;
3079         int i, num_eth_ports = 0;
3080         bool do_bond = true;
3081         struct mlx4_en_priv *priv;
3082         u8 v2p_port1 = 0;
3083         u8 v2p_port2 = 0;
3084
3085         if (!net_eq(dev_net(ndev), &init_net))
3086                 return NOTIFY_DONE;
3087
3088         mdev = container_of(this, struct mlx4_en_dev, nb);
3089         dev = mdev->dev;
3090
3091         /* Go into this mode only when two network devices set on two ports
3092          * of the same mlx4 device are slaves of the same bonding master
3093          */
3094         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
3095                 ++num_eth_ports;
3096                 if (!port && (mdev->pndev[i] == ndev))
3097                         port = i;
3098                 mdev->upper[i] = mdev->pndev[i] ?
3099                         netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
3100                 /* condition not met: network device is a slave */
3101                 if (!mdev->upper[i])
3102                         do_bond = false;
3103                 if (num_eth_ports < 2)
3104                         continue;
3105                 /* condition not met: same master */
3106                 if (mdev->upper[i] != mdev->upper[i-1])
3107                         do_bond = false;
3108         }
3109         /* condition not met: 2 salves */
3110         do_bond = (num_eth_ports ==  2) ? do_bond : false;
3111
3112         /* handle only events that come with enough info */
3113         if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
3114                 return NOTIFY_DONE;
3115
3116         priv = netdev_priv(ndev);
3117         if (do_bond) {
3118                 struct netdev_notifier_bonding_info *notifier_info = ptr;
3119                 struct netdev_bonding_info *bonding_info =
3120                         &notifier_info->bonding_info;
3121
3122                 /* required mode 1, 2 or 4 */
3123                 if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3124                     (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
3125                     (bonding_info->master.bond_mode != BOND_MODE_8023AD))
3126                         do_bond = false;
3127
3128                 /* require exactly 2 slaves */
3129                 if (bonding_info->master.num_slaves != 2)
3130                         do_bond = false;
3131
3132                 /* calc v2p */
3133                 if (do_bond) {
3134                         if (bonding_info->master.bond_mode ==
3135                             BOND_MODE_ACTIVEBACKUP) {
3136                                 /* in active-backup mode virtual ports are
3137                                  * mapped to the physical port of the active
3138                                  * slave */
3139                                 if (bonding_info->slave.state ==
3140                                     BOND_STATE_BACKUP) {
3141                                         if (port == 1) {
3142                                                 v2p_port1 = 2;
3143                                                 v2p_port2 = 2;
3144                                         } else {
3145                                                 v2p_port1 = 1;
3146                                                 v2p_port2 = 1;
3147                                         }
3148                                 } else { /* BOND_STATE_ACTIVE */
3149                                         if (port == 1) {
3150                                                 v2p_port1 = 1;
3151                                                 v2p_port2 = 1;
3152                                         } else {
3153                                                 v2p_port1 = 2;
3154                                                 v2p_port2 = 2;
3155                                         }
3156                                 }
3157                         } else { /* Active-Active */
3158                                 /* in active-active mode a virtual port is
3159                                  * mapped to the native physical port if and only
3160                                  * if the physical port is up */
3161                                 __s8 link = bonding_info->slave.link;
3162
3163                                 if (port == 1)
3164                                         v2p_port2 = 2;
3165                                 else
3166                                         v2p_port1 = 1;
3167                                 if ((link == BOND_LINK_UP) ||
3168                                     (link == BOND_LINK_FAIL)) {
3169                                         if (port == 1)
3170                                                 v2p_port1 = 1;
3171                                         else
3172                                                 v2p_port2 = 2;
3173                                 } else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3174                                         if (port == 1)
3175                                                 v2p_port1 = 2;
3176                                         else
3177                                                 v2p_port2 = 1;
3178                                 }
3179                         }
3180                 }
3181         }
3182
3183         mlx4_en_queue_bond_work(priv, do_bond,
3184                                 v2p_port1, v2p_port2);
3185
3186         return NOTIFY_DONE;
3187 }
3188
3189 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3190                                      struct mlx4_en_stats_bitmap *stats_bitmap,
3191                                      u8 rx_ppp, u8 rx_pause,
3192                                      u8 tx_ppp, u8 tx_pause)
3193 {
3194         int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3195
3196         if (!mlx4_is_slave(dev) &&
3197             (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3198                 mutex_lock(&stats_bitmap->mutex);
3199                 bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3200
3201                 if (rx_ppp)
3202                         bitmap_set(stats_bitmap->bitmap, last_i,
3203                                    NUM_FLOW_PRIORITY_STATS_RX);
3204                 last_i += NUM_FLOW_PRIORITY_STATS_RX;
3205
3206                 if (rx_pause && !(rx_ppp))
3207                         bitmap_set(stats_bitmap->bitmap, last_i,
3208                                    NUM_FLOW_STATS_RX);
3209                 last_i += NUM_FLOW_STATS_RX;
3210
3211                 if (tx_ppp)
3212                         bitmap_set(stats_bitmap->bitmap, last_i,
3213                                    NUM_FLOW_PRIORITY_STATS_TX);
3214                 last_i += NUM_FLOW_PRIORITY_STATS_TX;
3215
3216                 if (tx_pause && !(tx_ppp))
3217                         bitmap_set(stats_bitmap->bitmap, last_i,
3218                                    NUM_FLOW_STATS_TX);
3219                 last_i += NUM_FLOW_STATS_TX;
3220
3221                 mutex_unlock(&stats_bitmap->mutex);
3222         }
3223 }
3224
3225 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3226                               struct mlx4_en_stats_bitmap *stats_bitmap,
3227                               u8 rx_ppp, u8 rx_pause,
3228                               u8 tx_ppp, u8 tx_pause)
3229 {
3230         int last_i = 0;
3231
3232         mutex_init(&stats_bitmap->mutex);
3233         bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3234
3235         if (mlx4_is_slave(dev)) {
3236                 bitmap_set(stats_bitmap->bitmap, last_i +
3237                                          MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3238                 bitmap_set(stats_bitmap->bitmap, last_i +
3239                                          MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3240                 bitmap_set(stats_bitmap->bitmap, last_i +
3241                                          MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3242                 bitmap_set(stats_bitmap->bitmap, last_i +
3243                                          MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3244                 bitmap_set(stats_bitmap->bitmap, last_i +
3245                                          MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3246                 bitmap_set(stats_bitmap->bitmap, last_i +
3247                                          MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3248         } else {
3249                 bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3250         }
3251         last_i += NUM_MAIN_STATS;
3252
3253         bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3254         last_i += NUM_PORT_STATS;
3255
3256         if (mlx4_is_master(dev))
3257                 bitmap_set(stats_bitmap->bitmap, last_i,
3258                            NUM_PF_STATS);
3259         last_i += NUM_PF_STATS;
3260
3261         mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3262                                         rx_ppp, rx_pause,
3263                                         tx_ppp, tx_pause);
3264         last_i += NUM_FLOW_STATS;
3265
3266         if (!mlx4_is_slave(dev))
3267                 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3268         last_i += NUM_PKT_STATS;
3269
3270         bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3271         last_i += NUM_XDP_STATS;
3272 }
3273
3274 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3275                         struct mlx4_en_port_profile *prof)
3276 {
3277         struct net_device *dev;
3278         struct mlx4_en_priv *priv;
3279         int i, t;
3280         int err;
3281
3282         dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3283                                  MAX_TX_RINGS, MAX_RX_RINGS);
3284         if (dev == NULL)
3285                 return -ENOMEM;
3286
3287         netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3288         netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3289
3290         SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3291         dev->dev_port = port - 1;
3292
3293         /*
3294          * Initialize driver private data
3295          */
3296
3297         priv = netdev_priv(dev);
3298         memset(priv, 0, sizeof(struct mlx4_en_priv));
3299         priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3300         spin_lock_init(&priv->stats_lock);
3301         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3302         INIT_WORK(&priv->restart_task, mlx4_en_restart);
3303         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3304         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3305         INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3306         INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
3307         INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
3308 #ifdef CONFIG_RFS_ACCEL
3309         INIT_LIST_HEAD(&priv->filters);
3310         spin_lock_init(&priv->filters_lock);
3311 #endif
3312
3313         priv->dev = dev;
3314         priv->mdev = mdev;
3315         priv->ddev = &mdev->pdev->dev;
3316         priv->prof = prof;
3317         priv->port = port;
3318         priv->port_up = false;
3319         priv->flags = prof->flags;
3320         priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3321         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3322                         MLX4_WQE_CTRL_SOLICITED);
3323         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
3324         priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3325         netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3326
3327         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3328                 priv->tx_ring_num[t] = prof->tx_ring_num[t];
3329                 if (!priv->tx_ring_num[t])
3330                         continue;
3331
3332                 priv->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
3333                                            MAX_TX_RINGS, GFP_KERNEL);
3334                 if (!priv->tx_ring[t]) {
3335                         err = -ENOMEM;
3336                         goto out;
3337                 }
3338                 priv->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
3339                                          MAX_TX_RINGS, GFP_KERNEL);
3340                 if (!priv->tx_cq[t]) {
3341                         err = -ENOMEM;
3342                         goto out;
3343                 }
3344         }
3345         priv->rx_ring_num = prof->rx_ring_num;
3346         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3347         priv->cqe_size = mdev->dev->caps.cqe_size;
3348         priv->mac_index = -1;
3349         priv->msg_enable = MLX4_EN_MSG_LEVEL;
3350 #ifdef CONFIG_MLX4_EN_DCB
3351         if (!mlx4_is_slave(priv->mdev->dev)) {
3352                 u8 prio;
3353
3354                 for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; ++prio) {
3355                         priv->ets.prio_tc[prio] = prio;
3356                         priv->ets.tc_tsa[prio]  = IEEE_8021QAZ_TSA_VENDOR;
3357                 }
3358
3359                 priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3360                         DCB_CAP_DCBX_VER_IEEE;
3361                 priv->flags |= MLX4_EN_DCB_ENABLED;
3362                 priv->cee_config.pfc_state = false;
3363
3364                 for (i = 0; i < MLX4_EN_NUM_UP_HIGH; i++)
3365                         priv->cee_config.dcb_pfc[i] = pfc_disabled;
3366
3367                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3368                         dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3369                 } else {
3370                         en_info(priv, "enabling only PFC DCB ops\n");
3371                         dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3372                 }
3373         }
3374 #endif
3375
3376         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3377                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
3378
3379         /* Query for default mac and max mtu */
3380         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3381
3382         if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3383             MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3384                 priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3385
3386         /* Set default MAC */
3387         dev->addr_len = ETH_ALEN;
3388         mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3389         if (!is_valid_ether_addr(dev->dev_addr)) {
3390                 en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
3391                        priv->port, dev->dev_addr);
3392                 err = -EINVAL;
3393                 goto out;
3394         } else if (mlx4_is_slave(priv->mdev->dev) &&
3395                    (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3396                 /* Random MAC was assigned in mlx4_slave_cap
3397                  * in mlx4_core module
3398                  */
3399                 dev->addr_assign_type |= NET_ADDR_RANDOM;
3400                 en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3401         }
3402
3403         memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3404
3405         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3406                                           DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3407         err = mlx4_en_alloc_resources(priv);
3408         if (err)
3409                 goto out;
3410
3411         /* Initialize time stamping config */
3412         priv->hwtstamp_config.flags = 0;
3413         priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3414         priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3415
3416         /* Allocate page for receive rings */
3417         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3418                                 MLX4_EN_PAGE_SIZE);
3419         if (err) {
3420                 en_err(priv, "Failed to allocate page for rx qps\n");
3421                 goto out;
3422         }
3423         priv->allocated = 1;
3424
3425         /*
3426          * Initialize netdev entry points
3427          */
3428         if (mlx4_is_master(priv->mdev->dev))
3429                 dev->netdev_ops = &mlx4_netdev_ops_master;
3430         else
3431                 dev->netdev_ops = &mlx4_netdev_ops;
3432         dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3433         netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3434         netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3435
3436         dev->ethtool_ops = &mlx4_en_ethtool_ops;
3437
3438         /*
3439          * Set driver features
3440          */
3441         dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3442         if (mdev->LSO_support)
3443                 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3444
3445         dev->vlan_features = dev->hw_features;
3446
3447         dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3448         dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3449                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3450                         NETIF_F_HW_VLAN_CTAG_FILTER;
3451         dev->hw_features |= NETIF_F_LOOPBACK |
3452                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3453
3454         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3455                 dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3456                         NETIF_F_HW_VLAN_STAG_FILTER;
3457                 dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3458         }
3459
3460         if (mlx4_is_slave(mdev->dev)) {
3461                 bool vlan_offload_disabled;
3462                 int phv;
3463
3464                 err = get_phv_bit(mdev->dev, port, &phv);
3465                 if (!err && phv) {
3466                         dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3467                         priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3468                 }
3469                 err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3470                                                         &vlan_offload_disabled);
3471                 if (!err && vlan_offload_disabled) {
3472                         dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3473                                               NETIF_F_HW_VLAN_CTAG_RX |
3474                                               NETIF_F_HW_VLAN_STAG_TX |
3475                                               NETIF_F_HW_VLAN_STAG_RX);
3476                         dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3477                                            NETIF_F_HW_VLAN_CTAG_RX |
3478                                            NETIF_F_HW_VLAN_STAG_TX |
3479                                            NETIF_F_HW_VLAN_STAG_RX);
3480                 }
3481         } else {
3482                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3483                     !(mdev->dev->caps.flags2 &
3484                       MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3485                         dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3486         }
3487
3488         if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3489                 dev->hw_features |= NETIF_F_RXFCS;
3490
3491         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3492                 dev->hw_features |= NETIF_F_RXALL;
3493
3494         if (mdev->dev->caps.steering_mode ==
3495             MLX4_STEERING_MODE_DEVICE_MANAGED &&
3496             mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3497                 dev->hw_features |= NETIF_F_NTUPLE;
3498
3499         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3500                 dev->priv_flags |= IFF_UNICAST_FLT;
3501
3502         /* Setting a default hash function value */
3503         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3504                 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3505         } else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3506                 priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3507         } else {
3508                 en_warn(priv,
3509                         "No RSS hash capabilities exposed, using Toeplitz\n");
3510                 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3511         }
3512
3513         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3514                 dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3515                                     NETIF_F_GSO_UDP_TUNNEL_CSUM |
3516                                     NETIF_F_GSO_PARTIAL;
3517                 dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3518                                     NETIF_F_GSO_UDP_TUNNEL_CSUM |
3519                                     NETIF_F_GSO_PARTIAL;
3520                 dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3521         }
3522
3523         /* MTU range: 68 - hw-specific max */
3524         dev->min_mtu = ETH_MIN_MTU;
3525         dev->max_mtu = priv->max_mtu;
3526
3527         mdev->pndev[port] = dev;
3528         mdev->upper[port] = NULL;
3529
3530         netif_carrier_off(dev);
3531         mlx4_en_set_default_moderation(priv);
3532
3533         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3534         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3535
3536         mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3537
3538         /* Configure port */
3539         mlx4_en_calc_rx_buf(dev);
3540         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3541                                     priv->rx_skb_size + ETH_FCS_LEN,
3542                                     prof->tx_pause, prof->tx_ppp,
3543                                     prof->rx_pause, prof->rx_ppp);
3544         if (err) {
3545                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3546                        priv->port, err);
3547                 goto out;
3548         }
3549
3550         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3551                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3552                 if (err) {
3553                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3554                                err);
3555                         goto out;
3556                 }
3557         }
3558
3559         /* Init port */
3560         en_warn(priv, "Initializing port\n");
3561         err = mlx4_INIT_PORT(mdev->dev, priv->port);
3562         if (err) {
3563                 en_err(priv, "Failed Initializing port\n");
3564                 goto out;
3565         }
3566         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3567
3568         /* Initialize time stamp mechanism */
3569         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3570                 mlx4_en_init_timestamp(mdev);
3571
3572         queue_delayed_work(mdev->workqueue, &priv->service_task,
3573                            SERVICE_TASK_DELAY);
3574
3575         mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3576                                  mdev->profile.prof[priv->port].rx_ppp,
3577                                  mdev->profile.prof[priv->port].rx_pause,
3578                                  mdev->profile.prof[priv->port].tx_ppp,
3579                                  mdev->profile.prof[priv->port].tx_pause);
3580
3581         err = register_netdev(dev);
3582         if (err) {
3583                 en_err(priv, "Netdev registration failed for port %d\n", port);
3584                 goto out;
3585         }
3586
3587         priv->registered = 1;
3588         devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3589                                   dev);
3590
3591         return 0;
3592
3593 out:
3594         mlx4_en_destroy_netdev(dev);
3595         return err;
3596 }
3597
3598 int mlx4_en_reset_config(struct net_device *dev,
3599                          struct hwtstamp_config ts_config,
3600                          netdev_features_t features)
3601 {
3602         struct mlx4_en_priv *priv = netdev_priv(dev);
3603         struct mlx4_en_dev *mdev = priv->mdev;
3604         struct mlx4_en_port_profile new_prof;
3605         struct mlx4_en_priv *tmp;
3606         int port_up = 0;
3607         int err = 0;
3608
3609         if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3610             priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3611             !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3612             !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3613                 return 0; /* Nothing to change */
3614
3615         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3616             (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3617             (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3618                 en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3619                 return -EINVAL;
3620         }
3621
3622         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3623         if (!tmp)
3624                 return -ENOMEM;
3625
3626         mutex_lock(&mdev->state_lock);
3627
3628         memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3629         memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3630
3631         err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3632         if (err)
3633                 goto out;
3634
3635         if (priv->port_up) {
3636                 port_up = 1;
3637                 mlx4_en_stop_port(dev, 1);
3638         }
3639
3640         en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
3641                 ts_config.rx_filter,
3642                 !!(features & NETIF_F_HW_VLAN_CTAG_RX));
3643
3644         mlx4_en_safe_replace_resources(priv, tmp);
3645
3646         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3647                 if (features & NETIF_F_HW_VLAN_CTAG_RX)
3648                         dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3649                 else
3650                         dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3651         } else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3652                 /* RX time-stamping is OFF, update the RX vlan offload
3653                  * to the latest wanted state
3654                  */
3655                 if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3656                         dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3657                 else
3658                         dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3659         }
3660
3661         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3662                 if (features & NETIF_F_RXFCS)
3663                         dev->features |= NETIF_F_RXFCS;
3664                 else
3665                         dev->features &= ~NETIF_F_RXFCS;
3666         }
3667
3668         /* RX vlan offload and RX time-stamping can't co-exist !
3669          * Regardless of the caller's choice,
3670          * Turn Off RX vlan offload in case of time-stamping is ON
3671          */
3672         if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3673                 if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3674                         en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3675                 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3676         }
3677
3678         if (port_up) {
3679                 err = mlx4_en_start_port(dev);
3680                 if (err)
3681                         en_err(priv, "Failed starting port\n");
3682         }
3683
3684         if (!err)
3685                 err = mlx4_en_moderation_update(priv);
3686 out:
3687         mutex_unlock(&mdev->state_lock);
3688         kfree(tmp);
3689         if (!err)
3690                 netdev_features_change(dev);
3691         return err;
3692 }