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/etherdevice.h>
  35 #include <linux/tcp.h>
  36 #include <linux/if_vlan.h>
  37 #include <linux/delay.h>
  38 #include <linux/slab.h>
  39 #include <linux/hash.h>
  40 #include <net/ip.h>
  41 #include <net/busy_poll.h>
  42 #include <net/vxlan.h>
  43
  44 #include <linux/mlx4/driver.h>
  45 #include <linux/mlx4/device.h>
  46 #include <linux/mlx4/cmd.h>
  47 #include <linux/mlx4/cq.h>
  48
  49 #include "mlx4_en.h"
  50 #include "en_port.h"
  51
  52 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
  53 {
  54         struct mlx4_en_priv *priv = netdev_priv(dev);
  55         int i;
  56         unsigned int offset = 0;
  57
  58         if (up && up != MLX4_EN_NUM_UP)
  59                 return -EINVAL;
  60
  61         netdev_set_num_tc(dev, up);
  62
  63         /* Partition Tx queues evenly amongst UP's */
  64         for (i = 0; i < up; i++) {
  65                 netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
  66                 offset += priv->num_tx_rings_p_up;
  67         }
  68
  69         return 0;
  70 }
  71
  72 #ifdef CONFIG_NET_RX_BUSY_POLL
  73 /* must be called with local_bh_disable()d */
  74 static int mlx4_en_low_latency_recv(struct napi_struct *napi)
  75 {
  76         struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
  77         struct net_device *dev = cq->dev;
  78         struct mlx4_en_priv *priv = netdev_priv(dev);
  79         struct mlx4_en_rx_ring *rx_ring = priv->rx_ring[cq->ring];
  80         int done;
  81
  82         if (!priv->port_up)
  83                 return LL_FLUSH_FAILED;
  84
  85         if (!mlx4_en_cq_lock_poll(cq))
  86                 return LL_FLUSH_BUSY;
  87
  88         done = mlx4_en_process_rx_cq(dev, cq, 4);
  89         if (likely(done))
  90                 rx_ring->cleaned += done;
  91         else
  92                 rx_ring->misses++;
  93
  94         mlx4_en_cq_unlock_poll(cq);
  95
  96         return done;
  97 }
  98 #endif  /* CONFIG_NET_RX_BUSY_POLL */
  99
 100 #ifdef CONFIG_RFS_ACCEL
 101
 102 struct mlx4_en_filter {
 103         struct list_head next;
 104         struct work_struct work;
 105
 106         u8     ip_proto;
 107         __be32 src_ip;
 108         __be32 dst_ip;
 109         __be16 src_port;
 110         __be16 dst_port;
 111
 112         int rxq_index;
 113         struct mlx4_en_priv *priv;
 114         u32 flow_id;                    /* RFS infrastructure id */
 115         int id;                         /* mlx4_en driver id */
 116         u64 reg_id;                     /* Flow steering API id */
 117         u8 activated;                   /* Used to prevent expiry before filter
 118                                          * is attached
 119                                          */
 120         struct hlist_node filter_chain;
 121 };
 122
 123 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
 124
 125 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
 126 {
 127         switch (ip_proto) {
 128         case IPPROTO_UDP:
 129                 return MLX4_NET_TRANS_RULE_ID_UDP;
 130         case IPPROTO_TCP:
 131                 return MLX4_NET_TRANS_RULE_ID_TCP;
 132         default:
 133                 return MLX4_NET_TRANS_RULE_NUM;
 134         }
 135 };
 136
 137 static void mlx4_en_filter_work(struct work_struct *work)
 138 {
 139         struct mlx4_en_filter *filter = container_of(work,
 140                                                      struct mlx4_en_filter,
 141                                                      work);
 142         struct mlx4_en_priv *priv = filter->priv;
 143         struct mlx4_spec_list spec_tcp_udp = {
 144                 .id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
 145                 {
 146                         .tcp_udp = {
 147                                 .dst_port = filter->dst_port,
 148                                 .dst_port_msk = (__force __be16)-1,
 149                                 .src_port = filter->src_port,
 150                                 .src_port_msk = (__force __be16)-1,
 151                         },
 152                 },
 153         };
 154         struct mlx4_spec_list spec_ip = {
 155                 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
 156                 {
 157                         .ipv4 = {
 158                                 .dst_ip = filter->dst_ip,
 159                                 .dst_ip_msk = (__force __be32)-1,
 160                                 .src_ip = filter->src_ip,
 161                                 .src_ip_msk = (__force __be32)-1,
 162                         },
 163                 },
 164         };
 165         struct mlx4_spec_list spec_eth = {
 166                 .id = MLX4_NET_TRANS_RULE_ID_ETH,
 167         };
 168         struct mlx4_net_trans_rule rule = {
 169                 .list = LIST_HEAD_INIT(rule.list),
 170                 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
 171                 .exclusive = 1,
 172                 .allow_loopback = 1,
 173                 .promisc_mode = MLX4_FS_REGULAR,
 174                 .port = priv->port,
 175                 .priority = MLX4_DOMAIN_RFS,
 176         };
 177         int rc;
 178         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 179
 180         if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
 181                 en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
 182                         filter->ip_proto);
 183                 goto ignore;
 184         }
 185         list_add_tail(&spec_eth.list, &rule.list);
 186         list_add_tail(&spec_ip.list, &rule.list);
 187         list_add_tail(&spec_tcp_udp.list, &rule.list);
 188
 189         rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
 190         memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
 191         memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 192
 193         filter->activated = 0;
 194
 195         if (filter->reg_id) {
 196                 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 197                 if (rc && rc != -ENOENT)
 198                         en_err(priv, "Error detaching flow. rc = %d\n", rc);
 199         }
 200
 201         rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
 202         if (rc)
 203                 en_err(priv, "Error attaching flow. err = %d\n", rc);
 204
 205 ignore:
 206         mlx4_en_filter_rfs_expire(priv);
 207
 208         filter->activated = 1;
 209 }
 210
 211 static inline struct hlist_head *
 212 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 213                    __be16 src_port, __be16 dst_port)
 214 {
 215         unsigned long l;
 216         int bucket_idx;
 217
 218         l = (__force unsigned long)src_port |
 219             ((__force unsigned long)dst_port << 2);
 220         l ^= (__force unsigned long)(src_ip ^ dst_ip);
 221
 222         bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
 223
 224         return &priv->filter_hash[bucket_idx];
 225 }
 226
 227 static struct mlx4_en_filter *
 228 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
 229                      __be32 dst_ip, u8 ip_proto, __be16 src_port,
 230                      __be16 dst_port, u32 flow_id)
 231 {
 232         struct mlx4_en_filter *filter = NULL;
 233
 234         filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
 235         if (!filter)
 236                 return NULL;
 237
 238         filter->priv = priv;
 239         filter->rxq_index = rxq_index;
 240         INIT_WORK(&filter->work, mlx4_en_filter_work);
 241
 242         filter->src_ip = src_ip;
 243         filter->dst_ip = dst_ip;
 244         filter->ip_proto = ip_proto;
 245         filter->src_port = src_port;
 246         filter->dst_port = dst_port;
 247
 248         filter->flow_id = flow_id;
 249
 250         filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
 251
 252         list_add_tail(&filter->next, &priv->filters);
 253         hlist_add_head(&filter->filter_chain,
 254                        filter_hash_bucket(priv, src_ip, dst_ip, src_port,
 255                                           dst_port));
 256
 257         return filter;
 258 }
 259
 260 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
 261 {
 262         struct mlx4_en_priv *priv = filter->priv;
 263         int rc;
 264
 265         list_del(&filter->next);
 266
 267         rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 268         if (rc && rc != -ENOENT)
 269                 en_err(priv, "Error detaching flow. rc = %d\n", rc);
 270
 271         kfree(filter);
 272 }
 273
 274 static inline struct mlx4_en_filter *
 275 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 276                     u8 ip_proto, __be16 src_port, __be16 dst_port)
 277 {
 278         struct mlx4_en_filter *filter;
 279         struct mlx4_en_filter *ret = NULL;
 280
 281         hlist_for_each_entry(filter,
 282                              filter_hash_bucket(priv, src_ip, dst_ip,
 283                                                 src_port, dst_port),
 284                              filter_chain) {
 285                 if (filter->src_ip == src_ip &&
 286                     filter->dst_ip == dst_ip &&
 287                     filter->ip_proto == ip_proto &&
 288                     filter->src_port == src_port &&
 289                     filter->dst_port == dst_port) {
 290                         ret = filter;
 291                         break;
 292                 }
 293         }
 294
 295         return ret;
 296 }
 297
 298 static int
 299 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
 300                    u16 rxq_index, u32 flow_id)
 301 {
 302         struct mlx4_en_priv *priv = netdev_priv(net_dev);
 303         struct mlx4_en_filter *filter;
 304         const struct iphdr *ip;
 305         const __be16 *ports;
 306         u8 ip_proto;
 307         __be32 src_ip;
 308         __be32 dst_ip;
 309         __be16 src_port;
 310         __be16 dst_port;
 311         int nhoff = skb_network_offset(skb);
 312         int ret = 0;
 313
 314         if (skb->protocol != htons(ETH_P_IP))
 315                 return -EPROTONOSUPPORT;
 316
 317         ip = (const struct iphdr *)(skb->data + nhoff);
 318         if (ip_is_fragment(ip))
 319                 return -EPROTONOSUPPORT;
 320
 321         if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
 322                 return -EPROTONOSUPPORT;
 323         ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
 324
 325         ip_proto = ip->protocol;
 326         src_ip = ip->saddr;
 327         dst_ip = ip->daddr;
 328         src_port = ports[0];
 329         dst_port = ports[1];
 330
 331         spin_lock_bh(&priv->filters_lock);
 332         filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
 333                                      src_port, dst_port);
 334         if (filter) {
 335                 if (filter->rxq_index == rxq_index)
 336                         goto out;
 337
 338                 filter->rxq_index = rxq_index;
 339         } else {
 340                 filter = mlx4_en_filter_alloc(priv, rxq_index,
 341                                               src_ip, dst_ip, ip_proto,
 342                                               src_port, dst_port, flow_id);
 343                 if (!filter) {
 344                         ret = -ENOMEM;
 345                         goto err;
 346                 }
 347         }
 348
 349         queue_work(priv->mdev->workqueue, &filter->work);
 350
 351 out:
 352         ret = filter->id;
 353 err:
 354         spin_unlock_bh(&priv->filters_lock);
 355
 356         return ret;
 357 }
 358
 359 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
 360 {
 361         struct mlx4_en_filter *filter, *tmp;
 362         LIST_HEAD(del_list);
 363
 364         spin_lock_bh(&priv->filters_lock);
 365         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 366                 list_move(&filter->next, &del_list);
 367                 hlist_del(&filter->filter_chain);
 368         }
 369         spin_unlock_bh(&priv->filters_lock);
 370
 371         list_for_each_entry_safe(filter, tmp, &del_list, next) {
 372                 cancel_work_sync(&filter->work);
 373                 mlx4_en_filter_free(filter);
 374         }
 375 }
 376
 377 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
 378 {
 379         struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
 380         LIST_HEAD(del_list);
 381         int i = 0;
 382
 383         spin_lock_bh(&priv->filters_lock);
 384         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 385                 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
 386                         break;
 387
 388                 if (filter->activated &&
 389                     !work_pending(&filter->work) &&
 390                     rps_may_expire_flow(priv->dev,
 391                                         filter->rxq_index, filter->flow_id,
 392                                         filter->id)) {
 393                         list_move(&filter->next, &del_list);
 394                         hlist_del(&filter->filter_chain);
 395                 } else
 396                         last_filter = filter;
 397
 398                 i++;
 399         }
 400
 401         if (last_filter && (&last_filter->next != priv->filters.next))
 402                 list_move(&priv->filters, &last_filter->next);
 403
 404         spin_unlock_bh(&priv->filters_lock);
 405
 406         list_for_each_entry_safe(filter, tmp, &del_list, next)
 407                 mlx4_en_filter_free(filter);
 408 }
 409 #endif
 410
 411 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
 412                                    __be16 proto, u16 vid)
 413 {
 414         struct mlx4_en_priv *priv = netdev_priv(dev);
 415         struct mlx4_en_dev *mdev = priv->mdev;
 416         int err;
 417         int idx;
 418
 419         en_dbg(HW, priv, "adding VLAN:%d\n", vid);
 420
 421         set_bit(vid, priv->active_vlans);
 422
 423         /* Add VID to port VLAN filter */
 424         mutex_lock(&mdev->state_lock);
 425         if (mdev->device_up && priv->port_up) {
 426                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 427                 if (err)
 428                         en_err(priv, "Failed configuring VLAN filter\n");
 429         }
 430         if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
 431                 en_dbg(HW, priv, "failed adding vlan %d\n", vid);
 432         mutex_unlock(&mdev->state_lock);
 433
 434         return 0;
 435 }
 436
 437 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
 438                                     __be16 proto, u16 vid)
 439 {
 440         struct mlx4_en_priv *priv = netdev_priv(dev);
 441         struct mlx4_en_dev *mdev = priv->mdev;
 442         int err;
 443
 444         en_dbg(HW, priv, "Killing VID:%d\n", vid);
 445
 446         clear_bit(vid, priv->active_vlans);
 447
 448         /* Remove VID from port VLAN filter */
 449         mutex_lock(&mdev->state_lock);
 450         mlx4_unregister_vlan(mdev->dev, priv->port, vid);
 451
 452         if (mdev->device_up && priv->port_up) {
 453                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 454                 if (err)
 455                         en_err(priv, "Failed configuring VLAN filter\n");
 456         }
 457         mutex_unlock(&mdev->state_lock);
 458
 459         return 0;
 460 }
 461
 462 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
 463 {
 464         int i;
 465         for (i = ETH_ALEN - 1; i >= 0; --i) {
 466                 dst_mac[i] = src_mac & 0xff;
 467                 src_mac >>= 8;
 468         }
 469         memset(&dst_mac[ETH_ALEN], 0, 2);
 470 }
 471
 472
 473 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
 474                                     int qpn, u64 *reg_id)
 475 {
 476         int err;
 477         struct mlx4_spec_list spec_eth_outer = { {NULL} };
 478         struct mlx4_spec_list spec_vxlan     = { {NULL} };
 479         struct mlx4_spec_list spec_eth_inner = { {NULL} };
 480
 481         struct mlx4_net_trans_rule rule = {
 482                 .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 483                 .exclusive = 0,
 484                 .allow_loopback = 1,
 485                 .promisc_mode = MLX4_FS_REGULAR,
 486                 .priority = MLX4_DOMAIN_NIC,
 487         };
 488
 489         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 490
 491         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
 492                 return 0; /* do nothing */
 493
 494         rule.port = priv->port;
 495         rule.qpn = qpn;
 496         INIT_LIST_HEAD(&rule.list);
 497
 498         spec_eth_outer.id = MLX4_NET_TRANS_RULE_ID_ETH;
 499         memcpy(spec_eth_outer.eth.dst_mac, addr, ETH_ALEN);
 500         memcpy(spec_eth_outer.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 501
 502         spec_vxlan.id = MLX4_NET_TRANS_RULE_ID_VXLAN;    /* any vxlan header */
 503         spec_eth_inner.id = MLX4_NET_TRANS_RULE_ID_ETH;  /* any inner eth header */
 504
 505         list_add_tail(&spec_eth_outer.list, &rule.list);
 506         list_add_tail(&spec_vxlan.list,     &rule.list);
 507         list_add_tail(&spec_eth_inner.list, &rule.list);
 508
 509         err = mlx4_flow_attach(priv->mdev->dev, &rule, reg_id);
 510         if (err) {
 511                 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
 512                 return err;
 513         }
 514         en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
 515         return 0;
 516 }
 517
 518
 519 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
 520                                 unsigned char *mac, int *qpn, u64 *reg_id)
 521 {
 522         struct mlx4_en_dev *mdev = priv->mdev;
 523         struct mlx4_dev *dev = mdev->dev;
 524         int err;
 525
 526         switch (dev->caps.steering_mode) {
 527         case MLX4_STEERING_MODE_B0: {
 528                 struct mlx4_qp qp;
 529                 u8 gid[16] = {0};
 530
 531                 qp.qpn = *qpn;
 532                 memcpy(&gid[10], mac, ETH_ALEN);
 533                 gid[5] = priv->port;
 534
 535                 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
 536                 break;
 537         }
 538         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 539                 struct mlx4_spec_list spec_eth = { {NULL} };
 540                 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 541
 542                 struct mlx4_net_trans_rule rule = {
 543                         .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 544                         .exclusive = 0,
 545                         .allow_loopback = 1,
 546                         .promisc_mode = MLX4_FS_REGULAR,
 547                         .priority = MLX4_DOMAIN_NIC,
 548                 };
 549
 550                 rule.port = priv->port;
 551                 rule.qpn = *qpn;
 552                 INIT_LIST_HEAD(&rule.list);
 553
 554                 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
 555                 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
 556                 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 557                 list_add_tail(&spec_eth.list, &rule.list);
 558
 559                 err = mlx4_flow_attach(dev, &rule, reg_id);
 560                 break;
 561         }
 562         default:
 563                 return -EINVAL;
 564         }
 565         if (err)
 566                 en_warn(priv, "Failed Attaching Unicast\n");
 567
 568         return err;
 569 }
 570
 571 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
 572                                      unsigned char *mac, int qpn, u64 reg_id)
 573 {
 574         struct mlx4_en_dev *mdev = priv->mdev;
 575         struct mlx4_dev *dev = mdev->dev;
 576
 577         switch (dev->caps.steering_mode) {
 578         case MLX4_STEERING_MODE_B0: {
 579                 struct mlx4_qp qp;
 580                 u8 gid[16] = {0};
 581
 582                 qp.qpn = qpn;
 583                 memcpy(&gid[10], mac, ETH_ALEN);
 584                 gid[5] = priv->port;
 585
 586                 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
 587                 break;
 588         }
 589         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 590                 mlx4_flow_detach(dev, reg_id);
 591                 break;
 592         }
 593         default:
 594                 en_err(priv, "Invalid steering mode.\n");
 595         }
 596 }
 597
 598 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
 599 {
 600         struct mlx4_en_dev *mdev = priv->mdev;
 601         struct mlx4_dev *dev = mdev->dev;
 602         struct mlx4_mac_entry *entry;
 603         int index = 0;
 604         int err = 0;
 605         u64 reg_id;
 606         int *qpn = &priv->base_qpn;
 607         u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 608
 609         en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
 610                priv->dev->dev_addr);
 611         index = mlx4_register_mac(dev, priv->port, mac);
 612         if (index < 0) {
 613                 err = index;
 614                 en_err(priv, "Failed adding MAC: %pM\n",
 615                        priv->dev->dev_addr);
 616                 return err;
 617         }
 618
 619         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 620                 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
 621                 *qpn = base_qpn + index;
 622                 return 0;
 623         }
 624
 625         err = mlx4_qp_reserve_range(dev, 1, 1, qpn);
 626         en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
 627         if (err) {
 628                 en_err(priv, "Failed to reserve qp for mac registration\n");
 629                 goto qp_err;
 630         }
 631
 632         err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
 633         if (err)
 634                 goto steer_err;
 635
 636         err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
 637                                        &priv->tunnel_reg_id);
 638         if (err)
 639                 goto tunnel_err;
 640
 641         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 642         if (!entry) {
 643                 err = -ENOMEM;
 644                 goto alloc_err;
 645         }
 646         memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
 647         memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
 648         entry->reg_id = reg_id;
 649
 650         hlist_add_head_rcu(&entry->hlist,
 651                            &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
 652
 653         return 0;
 654
 655 alloc_err:
 656         if (priv->tunnel_reg_id)
 657                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 658 tunnel_err:
 659         mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
 660
 661 steer_err:
 662         mlx4_qp_release_range(dev, *qpn, 1);
 663
 664 qp_err:
 665         mlx4_unregister_mac(dev, priv->port, mac);
 666         return err;
 667 }
 668
 669 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
 670 {
 671         struct mlx4_en_dev *mdev = priv->mdev;
 672         struct mlx4_dev *dev = mdev->dev;
 673         int qpn = priv->base_qpn;
 674         u64 mac;
 675
 676         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 677                 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                 struct mlx4_mac_entry *entry;
 683                 struct hlist_node *tmp;
 684                 struct hlist_head *bucket;
 685                 unsigned int i;
 686
 687                 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
 688                         bucket = &priv->mac_hash[i];
 689                         hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 690                                 mac = mlx4_mac_to_u64(entry->mac);
 691                                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 692                                        entry->mac);
 693                                 mlx4_en_uc_steer_release(priv, entry->mac,
 694                                                          qpn, entry->reg_id);
 695
 696                                 mlx4_unregister_mac(dev, priv->port, mac);
 697                                 hlist_del_rcu(&entry->hlist);
 698                                 kfree_rcu(entry, rcu);
 699                         }
 700                 }
 701
 702                 if (priv->tunnel_reg_id) {
 703                         mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 704                         priv->tunnel_reg_id = 0;
 705                 }
 706
 707                 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
 708                        priv->port, qpn);
 709                 mlx4_qp_release_range(dev, qpn, 1);
 710                 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
 711         }
 712 }
 713
 714 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
 715                                unsigned char *new_mac, unsigned char *prev_mac)
 716 {
 717         struct mlx4_en_dev *mdev = priv->mdev;
 718         struct mlx4_dev *dev = mdev->dev;
 719         int err = 0;
 720         u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
 721
 722         if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
 723                 struct hlist_head *bucket;
 724                 unsigned int mac_hash;
 725                 struct mlx4_mac_entry *entry;
 726                 struct hlist_node *tmp;
 727                 u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
 728
 729                 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
 730                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 731                         if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
 732                                 mlx4_en_uc_steer_release(priv, entry->mac,
 733                                                          qpn, entry->reg_id);
 734                                 mlx4_unregister_mac(dev, priv->port,
 735                                                     prev_mac_u64);
 736                                 hlist_del_rcu(&entry->hlist);
 737                                 synchronize_rcu();
 738                                 memcpy(entry->mac, new_mac, ETH_ALEN);
 739                                 entry->reg_id = 0;
 740                                 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
 741                                 hlist_add_head_rcu(&entry->hlist,
 742                                                    &priv->mac_hash[mac_hash]);
 743                                 mlx4_register_mac(dev, priv->port, new_mac_u64);
 744                                 err = mlx4_en_uc_steer_add(priv, new_mac,
 745                                                            &qpn,
 746                                                            &entry->reg_id);
 747                                 if (err)
 748                                         return err;
 749                                 if (priv->tunnel_reg_id) {
 750                                         mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 751                                         priv->tunnel_reg_id = 0;
 752                                 }
 753                                 err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
 754                                                                &priv->tunnel_reg_id);
 755                                 return err;
 756                         }
 757                 }
 758                 return -EINVAL;
 759         }
 760
 761         return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
 762 }
 763
 764 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
 765                               unsigned char new_mac[ETH_ALEN + 2])
 766 {
 767         int err = 0;
 768
 769         if (priv->port_up) {
 770                 /* Remove old MAC and insert the new one */
 771                 err = mlx4_en_replace_mac(priv, priv->base_qpn,
 772                                           new_mac, priv->current_mac);
 773                 if (err)
 774                         en_err(priv, "Failed changing HW MAC address\n");
 775         } else
 776                 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
 777
 778         if (!err)
 779                 memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
 780
 781         return err;
 782 }
 783
 784 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
 785 {
 786         struct mlx4_en_priv *priv = netdev_priv(dev);
 787         struct mlx4_en_dev *mdev = priv->mdev;
 788         struct sockaddr *saddr = addr;
 789         unsigned char new_mac[ETH_ALEN + 2];
 790         int err;
 791
 792         if (!is_valid_ether_addr(saddr->sa_data))
 793                 return -EADDRNOTAVAIL;
 794
 795         mutex_lock(&mdev->state_lock);
 796         memcpy(new_mac, saddr->sa_data, ETH_ALEN);
 797         err = mlx4_en_do_set_mac(priv, new_mac);
 798         if (!err)
 799                 memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
 800         mutex_unlock(&mdev->state_lock);
 801
 802         return err;
 803 }
 804
 805 static void mlx4_en_clear_list(struct net_device *dev)
 806 {
 807         struct mlx4_en_priv *priv = netdev_priv(dev);
 808         struct mlx4_en_mc_list *tmp, *mc_to_del;
 809
 810         list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
 811                 list_del(&mc_to_del->list);
 812                 kfree(mc_to_del);
 813         }
 814 }
 815
 816 static void mlx4_en_cache_mclist(struct net_device *dev)
 817 {
 818         struct mlx4_en_priv *priv = netdev_priv(dev);
 819         struct netdev_hw_addr *ha;
 820         struct mlx4_en_mc_list *tmp;
 821
 822         mlx4_en_clear_list(dev);
 823         netdev_for_each_mc_addr(ha, dev) {
 824                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 825                 if (!tmp) {
 826                         mlx4_en_clear_list(dev);
 827                         return;
 828                 }
 829                 memcpy(tmp->addr, ha->addr, ETH_ALEN);
 830                 list_add_tail(&tmp->list, &priv->mc_list);
 831         }
 832 }
 833
 834 static void update_mclist_flags(struct mlx4_en_priv *priv,
 835                                 struct list_head *dst,
 836                                 struct list_head *src)
 837 {
 838         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 839         bool found;
 840
 841         /* Find all the entries that should be removed from dst,
 842          * These are the entries that are not found in src
 843          */
 844         list_for_each_entry(dst_tmp, dst, list) {
 845                 found = false;
 846                 list_for_each_entry(src_tmp, src, list) {
 847                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 848                                 found = true;
 849                                 break;
 850                         }
 851                 }
 852                 if (!found)
 853                         dst_tmp->action = MCLIST_REM;
 854         }
 855
 856         /* Add entries that exist in src but not in dst
 857          * mark them as need to add
 858          */
 859         list_for_each_entry(src_tmp, src, list) {
 860                 found = false;
 861                 list_for_each_entry(dst_tmp, dst, list) {
 862                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 863                                 dst_tmp->action = MCLIST_NONE;
 864                                 found = true;
 865                                 break;
 866                         }
 867                 }
 868                 if (!found) {
 869                         new_mc = kmemdup(src_tmp,
 870                                          sizeof(struct mlx4_en_mc_list),
 871                                          GFP_KERNEL);
 872                         if (!new_mc)
 873                                 return;
 874
 875                         new_mc->action = MCLIST_ADD;
 876                         list_add_tail(&new_mc->list, dst);
 877                 }
 878         }
 879 }
 880
 881 static void mlx4_en_set_rx_mode(struct net_device *dev)
 882 {
 883         struct mlx4_en_priv *priv = netdev_priv(dev);
 884
 885         if (!priv->port_up)
 886                 return;
 887
 888         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 889 }
 890
 891 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 892                                      struct mlx4_en_dev *mdev)
 893 {
 894         int err = 0;
 895
 896         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 897                 if (netif_msg_rx_status(priv))
 898                         en_warn(priv, "Entering promiscuous mode\n");
 899                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 900
 901                 /* Enable promiscouos mode */
 902                 switch (mdev->dev->caps.steering_mode) {
 903                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 904                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 905                                                           priv->port,
 906                                                           priv->base_qpn,
 907                                                           MLX4_FS_ALL_DEFAULT);
 908                         if (err)
 909                                 en_err(priv, "Failed enabling promiscuous mode\n");
 910                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 911                         break;
 912
 913                 case MLX4_STEERING_MODE_B0:
 914                         err = mlx4_unicast_promisc_add(mdev->dev,
 915                                                        priv->base_qpn,
 916                                                        priv->port);
 917                         if (err)
 918                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 919
 920                         /* Add the default qp number as multicast
 921                          * promisc
 922                          */
 923                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 924                                 err = mlx4_multicast_promisc_add(mdev->dev,
 925                                                                  priv->base_qpn,
 926                                                                  priv->port);
 927                                 if (err)
 928                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 929                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 930                         }
 931                         break;
 932
 933                 case MLX4_STEERING_MODE_A0:
 934                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 935                                                      priv->port,
 936                                                      priv->base_qpn,
 937                                                      1);
 938                         if (err)
 939                                 en_err(priv, "Failed enabling promiscuous mode\n");
 940                         break;
 941                 }
 942
 943                 /* Disable port multicast filter (unconditionally) */
 944                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 945                                           0, MLX4_MCAST_DISABLE);
 946                 if (err)
 947                         en_err(priv, "Failed disabling multicast filter\n");
 948         }
 949 }
 950
 951 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 952                                        struct mlx4_en_dev *mdev)
 953 {
 954         int err = 0;
 955
 956         if (netif_msg_rx_status(priv))
 957                 en_warn(priv, "Leaving promiscuous mode\n");
 958         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 959
 960         /* Disable promiscouos mode */
 961         switch (mdev->dev->caps.steering_mode) {
 962         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 963                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 964                                                      priv->port,
 965                                                      MLX4_FS_ALL_DEFAULT);
 966                 if (err)
 967                         en_err(priv, "Failed disabling promiscuous mode\n");
 968                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 969                 break;
 970
 971         case MLX4_STEERING_MODE_B0:
 972                 err = mlx4_unicast_promisc_remove(mdev->dev,
 973                                                   priv->base_qpn,
 974                                                   priv->port);
 975                 if (err)
 976                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 977                 /* Disable Multicast promisc */
 978                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 979                         err = mlx4_multicast_promisc_remove(mdev->dev,
 980                                                             priv->base_qpn,
 981                                                             priv->port);
 982                         if (err)
 983                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 984                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 985                 }
 986                 break;
 987
 988         case MLX4_STEERING_MODE_A0:
 989                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 990                                              priv->port,
 991                                              priv->base_qpn, 0);
 992                 if (err)
 993                         en_err(priv, "Failed disabling promiscuous mode\n");
 994                 break;
 995         }
 996 }
 997
 998 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
 999                                  struct net_device *dev,
1000                                  struct mlx4_en_dev *mdev)
1001 {
1002         struct mlx4_en_mc_list *mclist, *tmp;
1003         u64 mcast_addr = 0;
1004         u8 mc_list[16] = {0};
1005         int err = 0;
1006
1007         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
1008         if (dev->flags & IFF_ALLMULTI) {
1009                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1010                                           0, MLX4_MCAST_DISABLE);
1011                 if (err)
1012                         en_err(priv, "Failed disabling multicast filter\n");
1013
1014                 /* Add the default qp number as multicast promisc */
1015                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1016                         switch (mdev->dev->caps.steering_mode) {
1017                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
1018                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
1019                                                                   priv->port,
1020                                                                   priv->base_qpn,
1021                                                                   MLX4_FS_MC_DEFAULT);
1022                                 break;
1023
1024                         case MLX4_STEERING_MODE_B0:
1025                                 err = mlx4_multicast_promisc_add(mdev->dev,
1026                                                                  priv->base_qpn,
1027                                                                  priv->port);
1028                                 break;
1029
1030                         case MLX4_STEERING_MODE_A0:
1031                                 break;
1032                         }
1033                         if (err)
1034                                 en_err(priv, "Failed entering multicast promisc mode\n");
1035                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1036                 }
1037         } else {
1038                 /* Disable Multicast promisc */
1039                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1040                         switch (mdev->dev->caps.steering_mode) {
1041                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
1042                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
1043                                                                      priv->port,
1044                                                                      MLX4_FS_MC_DEFAULT);
1045                                 break;
1046
1047                         case MLX4_STEERING_MODE_B0:
1048                                 err = mlx4_multicast_promisc_remove(mdev->dev,
1049                                                                     priv->base_qpn,
1050                                                                     priv->port);
1051                                 break;
1052
1053                         case MLX4_STEERING_MODE_A0:
1054                                 break;
1055                         }
1056                         if (err)
1057                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
1058                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1059                 }
1060
1061                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1062                                           0, MLX4_MCAST_DISABLE);
1063                 if (err)
1064                         en_err(priv, "Failed disabling multicast filter\n");
1065
1066                 /* Flush mcast filter and init it with broadcast address */
1067                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1068                                     1, MLX4_MCAST_CONFIG);
1069
1070                 /* Update multicast list - we cache all addresses so they won't
1071                  * change while HW is updated holding the command semaphor */
1072                 netif_addr_lock_bh(dev);
1073                 mlx4_en_cache_mclist(dev);
1074                 netif_addr_unlock_bh(dev);
1075                 list_for_each_entry(mclist, &priv->mc_list, list) {
1076                         mcast_addr = mlx4_mac_to_u64(mclist->addr);
1077                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1078                                             mcast_addr, 0, MLX4_MCAST_CONFIG);
1079                 }
1080                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1081                                           0, MLX4_MCAST_ENABLE);
1082                 if (err)
1083                         en_err(priv, "Failed enabling multicast filter\n");
1084
1085                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1086                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1087                         if (mclist->action == MCLIST_REM) {
1088                                 /* detach this address and delete from list */
1089                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1090                                 mc_list[5] = priv->port;
1091                                 err = mlx4_multicast_detach(mdev->dev,
1092                                                             &priv->rss_map.indir_qp,
1093                                                             mc_list,
1094                                                             MLX4_PROT_ETH,
1095                                                             mclist->reg_id);
1096                                 if (err)
1097                                         en_err(priv, "Fail to detach multicast address\n");
1098
1099                                 if (mclist->tunnel_reg_id) {
1100                                         err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1101                                         if (err)
1102                                                 en_err(priv, "Failed to detach multicast address\n");
1103                                 }
1104
1105                                 /* remove from list */
1106                                 list_del(&mclist->list);
1107                                 kfree(mclist);
1108                         } else if (mclist->action == MCLIST_ADD) {
1109                                 /* attach the address */
1110                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1111                                 /* needed for B0 steering support */
1112                                 mc_list[5] = priv->port;
1113                                 err = mlx4_multicast_attach(mdev->dev,
1114                                                             &priv->rss_map.indir_qp,
1115                                                             mc_list,
1116                                                             priv->port, 0,
1117                                                             MLX4_PROT_ETH,
1118                                                             &mclist->reg_id);
1119                                 if (err)
1120                                         en_err(priv, "Fail to attach multicast address\n");
1121
1122                                 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1123                                                                &mclist->tunnel_reg_id);
1124                                 if (err)
1125                                         en_err(priv, "Failed to attach multicast address\n");
1126                         }
1127                 }
1128         }
1129 }
1130
1131 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1132                                  struct net_device *dev,
1133                                  struct mlx4_en_dev *mdev)
1134 {
1135         struct netdev_hw_addr *ha;
1136         struct mlx4_mac_entry *entry;
1137         struct hlist_node *tmp;
1138         bool found;
1139         u64 mac;
1140         int err = 0;
1141         struct hlist_head *bucket;
1142         unsigned int i;
1143         int removed = 0;
1144         u32 prev_flags;
1145
1146         /* Note that we do not need to protect our mac_hash traversal with rcu,
1147          * since all modification code is protected by mdev->state_lock
1148          */
1149
1150         /* find what to remove */
1151         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1152                 bucket = &priv->mac_hash[i];
1153                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1154                         found = false;
1155                         netdev_for_each_uc_addr(ha, dev) {
1156                                 if (ether_addr_equal_64bits(entry->mac,
1157                                                             ha->addr)) {
1158                                         found = true;
1159                                         break;
1160                                 }
1161                         }
1162
1163                         /* MAC address of the port is not in uc list */
1164                         if (ether_addr_equal_64bits(entry->mac,
1165                                                     priv->current_mac))
1166                                 found = true;
1167
1168                         if (!found) {
1169                                 mac = mlx4_mac_to_u64(entry->mac);
1170                                 mlx4_en_uc_steer_release(priv, entry->mac,
1171                                                          priv->base_qpn,
1172                                                          entry->reg_id);
1173                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1174
1175                                 hlist_del_rcu(&entry->hlist);
1176                                 kfree_rcu(entry, rcu);
1177                                 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1178                                        entry->mac, priv->port);
1179                                 ++removed;
1180                         }
1181                 }
1182         }
1183
1184         /* if we didn't remove anything, there is no use in trying to add
1185          * again once we are in a forced promisc mode state
1186          */
1187         if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1188                 return;
1189
1190         prev_flags = priv->flags;
1191         priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1192
1193         /* find what to add */
1194         netdev_for_each_uc_addr(ha, dev) {
1195                 found = false;
1196                 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1197                 hlist_for_each_entry(entry, bucket, hlist) {
1198                         if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1199                                 found = true;
1200                                 break;
1201                         }
1202                 }
1203
1204                 if (!found) {
1205                         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1206                         if (!entry) {
1207                                 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1208                                        ha->addr, priv->port);
1209                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1210                                 break;
1211                         }
1212                         mac = mlx4_mac_to_u64(ha->addr);
1213                         memcpy(entry->mac, ha->addr, ETH_ALEN);
1214                         err = mlx4_register_mac(mdev->dev, priv->port, mac);
1215                         if (err < 0) {
1216                                 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1217                                        ha->addr, priv->port, err);
1218                                 kfree(entry);
1219                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1220                                 break;
1221                         }
1222                         err = mlx4_en_uc_steer_add(priv, ha->addr,
1223                                                    &priv->base_qpn,
1224                                                    &entry->reg_id);
1225                         if (err) {
1226                                 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1227                                        ha->addr, priv->port, err);
1228                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1229                                 kfree(entry);
1230                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1231                                 break;
1232                         } else {
1233                                 unsigned int mac_hash;
1234                                 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1235                                        ha->addr, priv->port);
1236                                 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1237                                 bucket = &priv->mac_hash[mac_hash];
1238                                 hlist_add_head_rcu(&entry->hlist, bucket);
1239                         }
1240                 }
1241         }
1242
1243         if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1244                 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1245                         priv->port);
1246         } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1247                 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1248                         priv->port);
1249         }
1250 }
1251
1252 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1253 {
1254         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1255                                                  rx_mode_task);
1256         struct mlx4_en_dev *mdev = priv->mdev;
1257         struct net_device *dev = priv->dev;
1258
1259         mutex_lock(&mdev->state_lock);
1260         if (!mdev->device_up) {
1261                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1262                 goto out;
1263         }
1264         if (!priv->port_up) {
1265                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1266                 goto out;
1267         }
1268
1269         if (!netif_carrier_ok(dev)) {
1270                 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1271                         if (priv->port_state.link_state) {
1272                                 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1273                                 netif_carrier_on(dev);
1274                                 en_dbg(LINK, priv, "Link Up\n");
1275                         }
1276                 }
1277         }
1278
1279         if (dev->priv_flags & IFF_UNICAST_FLT)
1280                 mlx4_en_do_uc_filter(priv, dev, mdev);
1281
1282         /* Promsicuous mode: disable all filters */
1283         if ((dev->flags & IFF_PROMISC) ||
1284             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1285                 mlx4_en_set_promisc_mode(priv, mdev);
1286                 goto out;
1287         }
1288
1289         /* Not in promiscuous mode */
1290         if (priv->flags & MLX4_EN_FLAG_PROMISC)
1291                 mlx4_en_clear_promisc_mode(priv, mdev);
1292
1293         mlx4_en_do_multicast(priv, dev, mdev);
1294 out:
1295         mutex_unlock(&mdev->state_lock);
1296 }
1297
1298 #ifdef CONFIG_NET_POLL_CONTROLLER
1299 static void mlx4_en_netpoll(struct net_device *dev)
1300 {
1301         struct mlx4_en_priv *priv = netdev_priv(dev);
1302         struct mlx4_en_cq *cq;
1303         int i;
1304
1305         for (i = 0; i < priv->rx_ring_num; i++) {
1306                 cq = priv->rx_cq[i];
1307                 napi_schedule(&cq->napi);
1308         }
1309 }
1310 #endif
1311
1312 static void mlx4_en_tx_timeout(struct net_device *dev)
1313 {
1314         struct mlx4_en_priv *priv = netdev_priv(dev);
1315         struct mlx4_en_dev *mdev = priv->mdev;
1316         int i;
1317
1318         if (netif_msg_timer(priv))
1319                 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1320
1321         for (i = 0; i < priv->tx_ring_num; i++) {
1322                 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1323                         continue;
1324                 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1325                         i, priv->tx_ring[i]->qpn, priv->tx_ring[i]->cqn,
1326                         priv->tx_ring[i]->cons, priv->tx_ring[i]->prod);
1327         }
1328
1329         priv->port_stats.tx_timeout++;
1330         en_dbg(DRV, priv, "Scheduling watchdog\n");
1331         queue_work(mdev->workqueue, &priv->watchdog_task);
1332 }
1333
1334
1335 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
1336 {
1337         struct mlx4_en_priv *priv = netdev_priv(dev);
1338
1339         spin_lock_bh(&priv->stats_lock);
1340         memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
1341         spin_unlock_bh(&priv->stats_lock);
1342
1343         return &priv->ret_stats;
1344 }
1345
1346 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1347 {
1348         struct mlx4_en_cq *cq;
1349         int i;
1350
1351         /* If we haven't received a specific coalescing setting
1352          * (module param), we set the moderation parameters as follows:
1353          * - moder_cnt is set to the number of mtu sized packets to
1354          *   satisfy our coalescing target.
1355          * - moder_time is set to a fixed value.
1356          */
1357         priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1358         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1359         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1360         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1361         en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1362                priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1363
1364         /* Setup cq moderation params */
1365         for (i = 0; i < priv->rx_ring_num; i++) {
1366                 cq = priv->rx_cq[i];
1367                 cq->moder_cnt = priv->rx_frames;
1368                 cq->moder_time = priv->rx_usecs;
1369                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1370                 priv->last_moder_packets[i] = 0;
1371                 priv->last_moder_bytes[i] = 0;
1372         }
1373
1374         for (i = 0; i < priv->tx_ring_num; i++) {
1375                 cq = priv->tx_cq[i];
1376                 cq->moder_cnt = priv->tx_frames;
1377                 cq->moder_time = priv->tx_usecs;
1378         }
1379
1380         /* Reset auto-moderation params */
1381         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1382         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1383         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1384         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1385         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1386         priv->adaptive_rx_coal = 1;
1387         priv->last_moder_jiffies = 0;
1388         priv->last_moder_tx_packets = 0;
1389 }
1390
1391 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1392 {
1393         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1394         struct mlx4_en_cq *cq;
1395         unsigned long packets;
1396         unsigned long rate;
1397         unsigned long avg_pkt_size;
1398         unsigned long rx_packets;
1399         unsigned long rx_bytes;
1400         unsigned long rx_pkt_diff;
1401         int moder_time;
1402         int ring, err;
1403
1404         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1405                 return;
1406
1407         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1408                 spin_lock_bh(&priv->stats_lock);
1409                 rx_packets = priv->rx_ring[ring]->packets;
1410                 rx_bytes = priv->rx_ring[ring]->bytes;
1411                 spin_unlock_bh(&priv->stats_lock);
1412
1413                 rx_pkt_diff = ((unsigned long) (rx_packets -
1414                                 priv->last_moder_packets[ring]));
1415                 packets = rx_pkt_diff;
1416                 rate = packets * HZ / period;
1417                 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1418                                 priv->last_moder_bytes[ring])) / packets : 0;
1419
1420                 /* Apply auto-moderation only when packet rate
1421                  * exceeds a rate that it matters */
1422                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1423                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1424                         if (rate < priv->pkt_rate_low)
1425                                 moder_time = priv->rx_usecs_low;
1426                         else if (rate > priv->pkt_rate_high)
1427                                 moder_time = priv->rx_usecs_high;
1428                         else
1429                                 moder_time = (rate - priv->pkt_rate_low) *
1430                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1431                                         (priv->pkt_rate_high - priv->pkt_rate_low) +
1432                                         priv->rx_usecs_low;
1433                 } else {
1434                         moder_time = priv->rx_usecs_low;
1435                 }
1436
1437                 if (moder_time != priv->last_moder_time[ring]) {
1438                         priv->last_moder_time[ring] = moder_time;
1439                         cq = priv->rx_cq[ring];
1440                         cq->moder_time = moder_time;
1441                         cq->moder_cnt = priv->rx_frames;
1442                         err = mlx4_en_set_cq_moder(priv, cq);
1443                         if (err)
1444                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1445                                        ring);
1446                 }
1447                 priv->last_moder_packets[ring] = rx_packets;
1448                 priv->last_moder_bytes[ring] = rx_bytes;
1449         }
1450
1451         priv->last_moder_jiffies = jiffies;
1452 }
1453
1454 static void mlx4_en_do_get_stats(struct work_struct *work)
1455 {
1456         struct delayed_work *delay = to_delayed_work(work);
1457         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1458                                                  stats_task);
1459         struct mlx4_en_dev *mdev = priv->mdev;
1460         int err;
1461
1462         mutex_lock(&mdev->state_lock);
1463         if (mdev->device_up) {
1464                 if (priv->port_up) {
1465                         err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1466                         if (err)
1467                                 en_dbg(HW, priv, "Could not update stats\n");
1468
1469                         mlx4_en_auto_moderation(priv);
1470                 }
1471
1472                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1473         }
1474         if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1475                 mlx4_en_do_set_mac(priv, priv->current_mac);
1476                 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1477         }
1478         mutex_unlock(&mdev->state_lock);
1479 }
1480
1481 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1482  * periodically
1483  */
1484 static void mlx4_en_service_task(struct work_struct *work)
1485 {
1486         struct delayed_work *delay = to_delayed_work(work);
1487         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1488                                                  service_task);
1489         struct mlx4_en_dev *mdev = priv->mdev;
1490
1491         mutex_lock(&mdev->state_lock);
1492         if (mdev->device_up) {
1493                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1494                         mlx4_en_ptp_overflow_check(mdev);
1495
1496                 queue_delayed_work(mdev->workqueue, &priv->service_task,
1497                                    SERVICE_TASK_DELAY);
1498         }
1499         mutex_unlock(&mdev->state_lock);
1500 }
1501
1502 static void mlx4_en_linkstate(struct work_struct *work)
1503 {
1504         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1505                                                  linkstate_task);
1506         struct mlx4_en_dev *mdev = priv->mdev;
1507         int linkstate = priv->link_state;
1508
1509         mutex_lock(&mdev->state_lock);
1510         /* If observable port state changed set carrier state and
1511          * report to system log */
1512         if (priv->last_link_state != linkstate) {
1513                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1514                         en_info(priv, "Link Down\n");
1515                         netif_carrier_off(priv->dev);
1516                 } else {
1517                         en_info(priv, "Link Up\n");
1518                         netif_carrier_on(priv->dev);
1519                 }
1520         }
1521         priv->last_link_state = linkstate;
1522         mutex_unlock(&mdev->state_lock);
1523 }
1524
1525 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1526 {
1527         struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1528         int numa_node = priv->mdev->dev->numa_node;
1529         int ret = 0;
1530
1531         if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1532                 return -ENOMEM;
1533
1534         ret = cpumask_set_cpu_local_first(ring_idx, numa_node,
1535                                           ring->affinity_mask);
1536         if (ret)
1537                 free_cpumask_var(ring->affinity_mask);
1538
1539         return ret;
1540 }
1541
1542 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1543 {
1544         free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1545 }
1546
1547 int mlx4_en_start_port(struct net_device *dev)
1548 {
1549         struct mlx4_en_priv *priv = netdev_priv(dev);
1550         struct mlx4_en_dev *mdev = priv->mdev;
1551         struct mlx4_en_cq *cq;
1552         struct mlx4_en_tx_ring *tx_ring;
1553         int rx_index = 0;
1554         int tx_index = 0;
1555         int err = 0;
1556         int i;
1557         int j;
1558         u8 mc_list[16] = {0};
1559
1560         if (priv->port_up) {
1561                 en_dbg(DRV, priv, "start port called while port already up\n");
1562                 return 0;
1563         }
1564
1565         INIT_LIST_HEAD(&priv->mc_list);
1566         INIT_LIST_HEAD(&priv->curr_list);
1567         INIT_LIST_HEAD(&priv->ethtool_list);
1568         memset(&priv->ethtool_rules[0], 0,
1569                sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1570
1571         /* Calculate Rx buf size */
1572         dev->mtu = min(dev->mtu, priv->max_mtu);
1573         mlx4_en_calc_rx_buf(dev);
1574         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1575
1576         /* Configure rx cq's and rings */
1577         err = mlx4_en_activate_rx_rings(priv);
1578         if (err) {
1579                 en_err(priv, "Failed to activate RX rings\n");
1580                 return err;
1581         }
1582         for (i = 0; i < priv->rx_ring_num; i++) {
1583                 cq = priv->rx_cq[i];
1584
1585                 mlx4_en_cq_init_lock(cq);
1586
1587                 err = mlx4_en_init_affinity_hint(priv, i);
1588                 if (err) {
1589                         en_err(priv, "Failed preparing IRQ affinity hint\n");
1590                         goto cq_err;
1591                 }
1592
1593                 err = mlx4_en_activate_cq(priv, cq, i);
1594                 if (err) {
1595                         en_err(priv, "Failed activating Rx CQ\n");
1596                         mlx4_en_free_affinity_hint(priv, i);
1597                         goto cq_err;
1598                 }
1599                 for (j = 0; j < cq->size; j++)
1600                         cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1601                 err = mlx4_en_set_cq_moder(priv, cq);
1602                 if (err) {
1603                         en_err(priv, "Failed setting cq moderation parameters\n");
1604                         mlx4_en_deactivate_cq(priv, cq);
1605                         mlx4_en_free_affinity_hint(priv, i);
1606                         goto cq_err;
1607                 }
1608                 mlx4_en_arm_cq(priv, cq);
1609                 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1610                 ++rx_index;
1611         }
1612
1613         /* Set qp number */
1614         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1615         err = mlx4_en_get_qp(priv);
1616         if (err) {
1617                 en_err(priv, "Failed getting eth qp\n");
1618                 goto cq_err;
1619         }
1620         mdev->mac_removed[priv->port] = 0;
1621
1622         err = mlx4_en_config_rss_steer(priv);
1623         if (err) {
1624                 en_err(priv, "Failed configuring rss steering\n");
1625                 goto mac_err;
1626         }
1627
1628         err = mlx4_en_create_drop_qp(priv);
1629         if (err)
1630                 goto rss_err;
1631
1632         /* Configure tx cq's and rings */
1633         for (i = 0; i < priv->tx_ring_num; i++) {
1634                 /* Configure cq */
1635                 cq = priv->tx_cq[i];
1636                 err = mlx4_en_activate_cq(priv, cq, i);
1637                 if (err) {
1638                         en_err(priv, "Failed allocating Tx CQ\n");
1639                         goto tx_err;
1640                 }
1641                 err = mlx4_en_set_cq_moder(priv, cq);
1642                 if (err) {
1643                         en_err(priv, "Failed setting cq moderation parameters\n");
1644                         mlx4_en_deactivate_cq(priv, cq);
1645                         goto tx_err;
1646                 }
1647                 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1648                 cq->buf->wqe_index = cpu_to_be16(0xffff);
1649
1650                 /* Configure ring */
1651                 tx_ring = priv->tx_ring[i];
1652                 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1653                         i / priv->num_tx_rings_p_up);
1654                 if (err) {
1655                         en_err(priv, "Failed allocating Tx ring\n");
1656                         mlx4_en_deactivate_cq(priv, cq);
1657                         goto tx_err;
1658                 }
1659                 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1660
1661                 /* Arm CQ for TX completions */
1662                 mlx4_en_arm_cq(priv, cq);
1663
1664                 /* Set initial ownership of all Tx TXBBs to SW (1) */
1665                 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1666                         *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1667                 ++tx_index;
1668         }
1669
1670         /* Configure port */
1671         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1672                                     priv->rx_skb_size + ETH_FCS_LEN,
1673                                     priv->prof->tx_pause,
1674                                     priv->prof->tx_ppp,
1675                                     priv->prof->rx_pause,
1676                                     priv->prof->rx_ppp);
1677         if (err) {
1678                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1679                        priv->port, err);
1680                 goto tx_err;
1681         }
1682         /* Set default qp number */
1683         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1684         if (err) {
1685                 en_err(priv, "Failed setting default qp numbers\n");
1686                 goto tx_err;
1687         }
1688
1689         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1690                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1691                 if (err) {
1692                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1693                                err);
1694                         goto tx_err;
1695                 }
1696         }
1697
1698         /* Init port */
1699         en_dbg(HW, priv, "Initializing port\n");
1700         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1701         if (err) {
1702                 en_err(priv, "Failed Initializing port\n");
1703                 goto tx_err;
1704         }
1705
1706         /* Attach rx QP to bradcast address */
1707         memset(&mc_list[10], 0xff, ETH_ALEN);
1708         mc_list[5] = priv->port; /* needed for B0 steering support */
1709         if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1710                                   priv->port, 0, MLX4_PROT_ETH,
1711                                   &priv->broadcast_id))
1712                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1713
1714         /* Must redo promiscuous mode setup. */
1715         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1716
1717         /* Schedule multicast task to populate multicast list */
1718         queue_work(mdev->workqueue, &priv->rx_mode_task);
1719
1720         mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
1721
1722 #ifdef CONFIG_MLX4_EN_VXLAN
1723         if (priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1724                 vxlan_get_rx_port(dev);
1725 #endif
1726         priv->port_up = true;
1727         netif_tx_start_all_queues(dev);
1728         netif_device_attach(dev);
1729
1730         return 0;
1731
1732 tx_err:
1733         while (tx_index--) {
1734                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1735                 mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1736         }
1737         mlx4_en_destroy_drop_qp(priv);
1738 rss_err:
1739         mlx4_en_release_rss_steer(priv);
1740 mac_err:
1741         mlx4_en_put_qp(priv);
1742 cq_err:
1743         while (rx_index--) {
1744                 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1745                 mlx4_en_free_affinity_hint(priv, i);
1746         }
1747         for (i = 0; i < priv->rx_ring_num; i++)
1748                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1749
1750         return err; /* need to close devices */
1751 }
1752
1753
1754 void mlx4_en_stop_port(struct net_device *dev, int detach)
1755 {
1756         struct mlx4_en_priv *priv = netdev_priv(dev);
1757         struct mlx4_en_dev *mdev = priv->mdev;
1758         struct mlx4_en_mc_list *mclist, *tmp;
1759         struct ethtool_flow_id *flow, *tmp_flow;
1760         int i;
1761         u8 mc_list[16] = {0};
1762
1763         if (!priv->port_up) {
1764                 en_dbg(DRV, priv, "stop port called while port already down\n");
1765                 return;
1766         }
1767
1768         /* close port*/
1769         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1770
1771         /* Synchronize with tx routine */
1772         netif_tx_lock_bh(dev);
1773         if (detach)
1774                 netif_device_detach(dev);
1775         netif_tx_stop_all_queues(dev);
1776         netif_tx_unlock_bh(dev);
1777
1778         netif_tx_disable(dev);
1779
1780         /* Set port as not active */
1781         priv->port_up = false;
1782
1783         /* Promsicuous mode */
1784         if (mdev->dev->caps.steering_mode ==
1785             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1786                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1787                                  MLX4_EN_FLAG_MC_PROMISC);
1788                 mlx4_flow_steer_promisc_remove(mdev->dev,
1789                                                priv->port,
1790                                                MLX4_FS_ALL_DEFAULT);
1791                 mlx4_flow_steer_promisc_remove(mdev->dev,
1792                                                priv->port,
1793                                                MLX4_FS_MC_DEFAULT);
1794         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1795                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1796
1797                 /* Disable promiscouos mode */
1798                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1799                                             priv->port);
1800
1801                 /* Disable Multicast promisc */
1802                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1803                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1804                                                       priv->port);
1805                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1806                 }
1807         }
1808
1809         /* Detach All multicasts */
1810         memset(&mc_list[10], 0xff, ETH_ALEN);
1811         mc_list[5] = priv->port; /* needed for B0 steering support */
1812         mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1813                               MLX4_PROT_ETH, priv->broadcast_id);
1814         list_for_each_entry(mclist, &priv->curr_list, list) {
1815                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1816                 mc_list[5] = priv->port;
1817                 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1818                                       mc_list, MLX4_PROT_ETH, mclist->reg_id);
1819                 if (mclist->tunnel_reg_id)
1820                         mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1821         }
1822         mlx4_en_clear_list(dev);
1823         list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1824                 list_del(&mclist->list);
1825                 kfree(mclist);
1826         }
1827
1828         /* Flush multicast filter */
1829         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1830
1831         /* Remove flow steering rules for the port*/
1832         if (mdev->dev->caps.steering_mode ==
1833             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1834                 ASSERT_RTNL();
1835                 list_for_each_entry_safe(flow, tmp_flow,
1836                                          &priv->ethtool_list, list) {
1837                         mlx4_flow_detach(mdev->dev, flow->id);
1838                         list_del(&flow->list);
1839                 }
1840         }
1841
1842         mlx4_en_destroy_drop_qp(priv);
1843
1844         /* Free TX Rings */
1845         for (i = 0; i < priv->tx_ring_num; i++) {
1846                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1847                 mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1848         }
1849         msleep(10);
1850
1851         for (i = 0; i < priv->tx_ring_num; i++)
1852                 mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1853
1854         /* Free RSS qps */
1855         mlx4_en_release_rss_steer(priv);
1856
1857         /* Unregister Mac address for the port */
1858         mlx4_en_put_qp(priv);
1859         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1860                 mdev->mac_removed[priv->port] = 1;
1861
1862         /* Free RX Rings */
1863         for (i = 0; i < priv->rx_ring_num; i++) {
1864                 struct mlx4_en_cq *cq = priv->rx_cq[i];
1865
1866                 local_bh_disable();
1867                 while (!mlx4_en_cq_lock_napi(cq)) {
1868                         pr_info("CQ %d locked\n", i);
1869                         mdelay(1);
1870                 }
1871                 local_bh_enable();
1872
1873                 while (test_bit(NAPI_STATE_SCHED, &cq->napi.state))
1874                         msleep(1);
1875                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1876                 mlx4_en_deactivate_cq(priv, cq);
1877
1878                 mlx4_en_free_affinity_hint(priv, i);
1879         }
1880 }
1881
1882 static void mlx4_en_restart(struct work_struct *work)
1883 {
1884         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1885                                                  watchdog_task);
1886         struct mlx4_en_dev *mdev = priv->mdev;
1887         struct net_device *dev = priv->dev;
1888
1889         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1890
1891         mutex_lock(&mdev->state_lock);
1892         if (priv->port_up) {
1893                 mlx4_en_stop_port(dev, 1);
1894                 if (mlx4_en_start_port(dev))
1895                         en_err(priv, "Failed restarting port %d\n", priv->port);
1896         }
1897         mutex_unlock(&mdev->state_lock);
1898 }
1899
1900 static void mlx4_en_clear_stats(struct net_device *dev)
1901 {
1902         struct mlx4_en_priv *priv = netdev_priv(dev);
1903         struct mlx4_en_dev *mdev = priv->mdev;
1904         int i;
1905
1906         if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1907                 en_dbg(HW, priv, "Failed dumping statistics\n");
1908
1909         memset(&priv->stats, 0, sizeof(priv->stats));
1910         memset(&priv->pstats, 0, sizeof(priv->pstats));
1911         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1912         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1913
1914         for (i = 0; i < priv->tx_ring_num; i++) {
1915                 priv->tx_ring[i]->bytes = 0;
1916                 priv->tx_ring[i]->packets = 0;
1917                 priv->tx_ring[i]->tx_csum = 0;
1918         }
1919         for (i = 0; i < priv->rx_ring_num; i++) {
1920                 priv->rx_ring[i]->bytes = 0;
1921                 priv->rx_ring[i]->packets = 0;
1922                 priv->rx_ring[i]->csum_ok = 0;
1923                 priv->rx_ring[i]->csum_none = 0;
1924         }
1925 }
1926
1927 static int mlx4_en_open(struct net_device *dev)
1928 {
1929         struct mlx4_en_priv *priv = netdev_priv(dev);
1930         struct mlx4_en_dev *mdev = priv->mdev;
1931         int err = 0;
1932
1933         mutex_lock(&mdev->state_lock);
1934
1935         if (!mdev->device_up) {
1936                 en_err(priv, "Cannot open - device down/disabled\n");
1937                 err = -EBUSY;
1938                 goto out;
1939         }
1940
1941         /* Reset HW statistics and SW counters */
1942         mlx4_en_clear_stats(dev);
1943
1944         err = mlx4_en_start_port(dev);
1945         if (err)
1946                 en_err(priv, "Failed starting port:%d\n", priv->port);
1947
1948 out:
1949         mutex_unlock(&mdev->state_lock);
1950         return err;
1951 }
1952
1953
1954 static int mlx4_en_close(struct net_device *dev)
1955 {
1956         struct mlx4_en_priv *priv = netdev_priv(dev);
1957         struct mlx4_en_dev *mdev = priv->mdev;
1958
1959         en_dbg(IFDOWN, priv, "Close port called\n");
1960
1961         mutex_lock(&mdev->state_lock);
1962
1963         mlx4_en_stop_port(dev, 0);
1964         netif_carrier_off(dev);
1965
1966         mutex_unlock(&mdev->state_lock);
1967         return 0;
1968 }
1969
1970 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1971 {
1972         int i;
1973
1974 #ifdef CONFIG_RFS_ACCEL
1975         free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1976         priv->dev->rx_cpu_rmap = NULL;
1977 #endif
1978
1979         for (i = 0; i < priv->tx_ring_num; i++) {
1980                 if (priv->tx_ring && priv->tx_ring[i])
1981                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1982                 if (priv->tx_cq && priv->tx_cq[i])
1983                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1984         }
1985
1986         for (i = 0; i < priv->rx_ring_num; i++) {
1987                 if (priv->rx_ring[i])
1988                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1989                                 priv->prof->rx_ring_size, priv->stride);
1990                 if (priv->rx_cq[i])
1991                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1992         }
1993
1994         if (priv->base_tx_qpn) {
1995                 mlx4_qp_release_range(priv->mdev->dev, priv->base_tx_qpn, priv->tx_ring_num);
1996                 priv->base_tx_qpn = 0;
1997         }
1998 }
1999
2000 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2001 {
2002         struct mlx4_en_port_profile *prof = priv->prof;
2003         int i;
2004         int err;
2005         int node;
2006
2007         err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &priv->base_tx_qpn);
2008         if (err) {
2009                 en_err(priv, "failed reserving range for TX rings\n");
2010                 return err;
2011         }
2012
2013         /* Create tx Rings */
2014         for (i = 0; i < priv->tx_ring_num; i++) {
2015                 node = cpu_to_node(i % num_online_cpus());
2016                 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
2017                                       prof->tx_ring_size, i, TX, node))
2018                         goto err;
2019
2020                 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
2021                                            priv->base_tx_qpn + i,
2022                                            prof->tx_ring_size, TXBB_SIZE,
2023                                            node, i))
2024                         goto err;
2025         }
2026
2027         /* Create rx Rings */
2028         for (i = 0; i < priv->rx_ring_num; i++) {
2029                 node = cpu_to_node(i % num_online_cpus());
2030                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2031                                       prof->rx_ring_size, i, RX, node))
2032                         goto err;
2033
2034                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2035                                            prof->rx_ring_size, priv->stride,
2036                                            node))
2037                         goto err;
2038         }
2039
2040 #ifdef CONFIG_RFS_ACCEL
2041         if (priv->mdev->dev->caps.comp_pool) {
2042                 priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->mdev->dev->caps.comp_pool);
2043                 if (!priv->dev->rx_cpu_rmap)
2044                         goto err;
2045         }
2046 #endif
2047
2048         return 0;
2049
2050 err:
2051         en_err(priv, "Failed to allocate NIC resources\n");
2052         for (i = 0; i < priv->rx_ring_num; i++) {
2053                 if (priv->rx_ring[i])
2054                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2055                                                 prof->rx_ring_size,
2056                                                 priv->stride);
2057                 if (priv->rx_cq[i])
2058                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2059         }
2060         for (i = 0; i < priv->tx_ring_num; i++) {
2061                 if (priv->tx_ring[i])
2062                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
2063                 if (priv->tx_cq[i])
2064                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
2065         }
2066         return -ENOMEM;
2067 }
2068
2069
2070 void mlx4_en_destroy_netdev(struct net_device *dev)
2071 {
2072         struct mlx4_en_priv *priv = netdev_priv(dev);
2073         struct mlx4_en_dev *mdev = priv->mdev;
2074
2075         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2076
2077         /* Unregister device - this will close the port if it was up */
2078         if (priv->registered)
2079                 unregister_netdev(dev);
2080
2081         if (priv->allocated)
2082                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2083
2084         cancel_delayed_work(&priv->stats_task);
2085         cancel_delayed_work(&priv->service_task);
2086         /* flush any pending task for this netdev */
2087         flush_workqueue(mdev->workqueue);
2088
2089         /* Detach the netdev so tasks would not attempt to access it */
2090         mutex_lock(&mdev->state_lock);
2091         mdev->pndev[priv->port] = NULL;
2092         mutex_unlock(&mdev->state_lock);
2093
2094         mlx4_en_free_resources(priv);
2095
2096         kfree(priv->tx_ring);
2097         kfree(priv->tx_cq);
2098
2099         free_netdev(dev);
2100 }
2101
2102 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2103 {
2104         struct mlx4_en_priv *priv = netdev_priv(dev);
2105         struct mlx4_en_dev *mdev = priv->mdev;
2106         int err = 0;
2107
2108         en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2109                  dev->mtu, new_mtu);
2110
2111         if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
2112                 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
2113                 return -EPERM;
2114         }
2115         dev->mtu = new_mtu;
2116
2117         if (netif_running(dev)) {
2118                 mutex_lock(&mdev->state_lock);
2119                 if (!mdev->device_up) {
2120                         /* NIC is probably restarting - let watchdog task reset
2121                          * the port */
2122                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2123                 } else {
2124                         mlx4_en_stop_port(dev, 1);
2125                         err = mlx4_en_start_port(dev);
2126                         if (err) {
2127                                 en_err(priv, "Failed restarting port:%d\n",
2128                                          priv->port);
2129                                 queue_work(mdev->workqueue, &priv->watchdog_task);
2130                         }
2131                 }
2132                 mutex_unlock(&mdev->state_lock);
2133         }
2134         return 0;
2135 }
2136
2137 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2138 {
2139         struct mlx4_en_priv *priv = netdev_priv(dev);
2140         struct mlx4_en_dev *mdev = priv->mdev;
2141         struct hwtstamp_config config;
2142
2143         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2144                 return -EFAULT;
2145
2146         /* reserved for future extensions */
2147         if (config.flags)
2148                 return -EINVAL;
2149
2150         /* device doesn't support time stamping */
2151         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2152                 return -EINVAL;
2153
2154         /* TX HW timestamp */
2155         switch (config.tx_type) {
2156         case HWTSTAMP_TX_OFF:
2157         case HWTSTAMP_TX_ON:
2158                 break;
2159         default:
2160                 return -ERANGE;
2161         }
2162
2163         /* RX HW timestamp */
2164         switch (config.rx_filter) {
2165         case HWTSTAMP_FILTER_NONE:
2166                 break;
2167         case HWTSTAMP_FILTER_ALL:
2168         case HWTSTAMP_FILTER_SOME:
2169         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2170         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2171         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2172         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2173         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2174         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2175         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2176         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2177         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2178         case HWTSTAMP_FILTER_PTP_V2_EVENT:
2179         case HWTSTAMP_FILTER_PTP_V2_SYNC:
2180         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2181                 config.rx_filter = HWTSTAMP_FILTER_ALL;
2182                 break;
2183         default:
2184                 return -ERANGE;
2185         }
2186
2187         if (mlx4_en_timestamp_config(dev, config.tx_type, config.rx_filter)) {
2188                 config.tx_type = HWTSTAMP_TX_OFF;
2189                 config.rx_filter = HWTSTAMP_FILTER_NONE;
2190         }
2191
2192         return copy_to_user(ifr->ifr_data, &config,
2193                             sizeof(config)) ? -EFAULT : 0;
2194 }
2195
2196 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2197 {
2198         struct mlx4_en_priv *priv = netdev_priv(dev);
2199
2200         return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2201                             sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2202 }
2203
2204 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2205 {
2206         switch (cmd) {
2207         case SIOCSHWTSTAMP:
2208                 return mlx4_en_hwtstamp_set(dev, ifr);
2209         case SIOCGHWTSTAMP:
2210                 return mlx4_en_hwtstamp_get(dev, ifr);
2211         default:
2212                 return -EOPNOTSUPP;
2213         }
2214 }
2215
2216 static int mlx4_en_set_features(struct net_device *netdev,
2217                 netdev_features_t features)
2218 {
2219         struct mlx4_en_priv *priv = netdev_priv(netdev);
2220
2221         if (features & NETIF_F_LOOPBACK)
2222                 priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
2223         else
2224                 priv->ctrl_flags &=
2225                         cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
2226
2227         mlx4_en_update_loopback_state(netdev, features);
2228
2229         return 0;
2230
2231 }
2232
2233 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2234 {
2235         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2236         struct mlx4_en_dev *mdev = en_priv->mdev;
2237         u64 mac_u64 = mlx4_mac_to_u64(mac);
2238
2239         if (!is_valid_ether_addr(mac))
2240                 return -EINVAL;
2241
2242         return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2243 }
2244
2245 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos)
2246 {
2247         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2248         struct mlx4_en_dev *mdev = en_priv->mdev;
2249
2250         return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos);
2251 }
2252
2253 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2254 {
2255         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2256         struct mlx4_en_dev *mdev = en_priv->mdev;
2257
2258         return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2259 }
2260
2261 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2262 {
2263         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2264         struct mlx4_en_dev *mdev = en_priv->mdev;
2265
2266         return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2267 }
2268
2269 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2270 {
2271         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2272         struct mlx4_en_dev *mdev = en_priv->mdev;
2273
2274         return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2275 }
2276
2277 #define PORT_ID_BYTE_LEN 8
2278 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2279                                     struct netdev_phys_port_id *ppid)
2280 {
2281         struct mlx4_en_priv *priv = netdev_priv(dev);
2282         struct mlx4_dev *mdev = priv->mdev->dev;
2283         int i;
2284         u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2285
2286         if (!phys_port_id)
2287                 return -EOPNOTSUPP;
2288
2289         ppid->id_len = sizeof(phys_port_id);
2290         for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2291                 ppid->id[i] =  phys_port_id & 0xff;
2292                 phys_port_id >>= 8;
2293         }
2294         return 0;
2295 }
2296
2297 #ifdef CONFIG_MLX4_EN_VXLAN
2298 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2299 {
2300         int ret;
2301         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2302                                                  vxlan_add_task);
2303
2304         ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2305         if (ret)
2306                 goto out;
2307
2308         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2309                                   VXLAN_STEER_BY_OUTER_MAC, 1);
2310 out:
2311         if (ret)
2312                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2313 }
2314
2315 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2316 {
2317         int ret;
2318         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2319                                                  vxlan_del_task);
2320
2321         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2322                                   VXLAN_STEER_BY_OUTER_MAC, 0);
2323         if (ret)
2324                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2325
2326         priv->vxlan_port = 0;
2327 }
2328
2329 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2330                                    sa_family_t sa_family, __be16 port)
2331 {
2332         struct mlx4_en_priv *priv = netdev_priv(dev);
2333         __be16 current_port;
2334
2335         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2336                 return;
2337
2338         if (sa_family == AF_INET6)
2339                 return;
2340
2341         current_port = priv->vxlan_port;
2342         if (current_port && current_port != port) {
2343                 en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2344                         ntohs(current_port), ntohs(port));
2345                 return;
2346         }
2347
2348         priv->vxlan_port = port;
2349         queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2350 }
2351
2352 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2353                                    sa_family_t sa_family, __be16 port)
2354 {
2355         struct mlx4_en_priv *priv = netdev_priv(dev);
2356         __be16 current_port;
2357
2358         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2359                 return;
2360
2361         if (sa_family == AF_INET6)
2362                 return;
2363
2364         current_port = priv->vxlan_port;
2365         if (current_port != port) {
2366                 en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2367                 return;
2368         }
2369
2370         queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2371 }
2372 #endif
2373
2374 static const struct net_device_ops mlx4_netdev_ops = {
2375         .ndo_open               = mlx4_en_open,
2376         .ndo_stop               = mlx4_en_close,
2377         .ndo_start_xmit         = mlx4_en_xmit,
2378         .ndo_select_queue       = mlx4_en_select_queue,
2379         .ndo_get_stats          = mlx4_en_get_stats,
2380         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2381         .ndo_set_mac_address    = mlx4_en_set_mac,
2382         .ndo_validate_addr      = eth_validate_addr,
2383         .ndo_change_mtu         = mlx4_en_change_mtu,
2384         .ndo_do_ioctl           = mlx4_en_ioctl,
2385         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2386         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2387         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2388 #ifdef CONFIG_NET_POLL_CONTROLLER
2389         .ndo_poll_controller    = mlx4_en_netpoll,
2390 #endif
2391         .ndo_set_features       = mlx4_en_set_features,
2392         .ndo_setup_tc           = mlx4_en_setup_tc,
2393 #ifdef CONFIG_RFS_ACCEL
2394         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2395 #endif
2396 #ifdef CONFIG_NET_RX_BUSY_POLL
2397         .ndo_busy_poll          = mlx4_en_low_latency_recv,
2398 #endif
2399         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2400 #ifdef CONFIG_MLX4_EN_VXLAN
2401         .ndo_add_vxlan_port     = mlx4_en_add_vxlan_port,
2402         .ndo_del_vxlan_port     = mlx4_en_del_vxlan_port,
2403 #endif
2404 };
2405
2406 static const struct net_device_ops mlx4_netdev_ops_master = {
2407         .ndo_open               = mlx4_en_open,
2408         .ndo_stop               = mlx4_en_close,
2409         .ndo_start_xmit         = mlx4_en_xmit,
2410         .ndo_select_queue       = mlx4_en_select_queue,
2411         .ndo_get_stats          = mlx4_en_get_stats,
2412         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2413         .ndo_set_mac_address    = mlx4_en_set_mac,
2414         .ndo_validate_addr      = eth_validate_addr,
2415         .ndo_change_mtu         = mlx4_en_change_mtu,
2416         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2417         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2418         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2419         .ndo_set_vf_mac         = mlx4_en_set_vf_mac,
2420         .ndo_set_vf_vlan        = mlx4_en_set_vf_vlan,
2421         .ndo_set_vf_spoofchk    = mlx4_en_set_vf_spoofchk,
2422         .ndo_set_vf_link_state  = mlx4_en_set_vf_link_state,
2423         .ndo_get_vf_config      = mlx4_en_get_vf_config,
2424 #ifdef CONFIG_NET_POLL_CONTROLLER
2425         .ndo_poll_controller    = mlx4_en_netpoll,
2426 #endif
2427         .ndo_set_features       = mlx4_en_set_features,
2428         .ndo_setup_tc           = mlx4_en_setup_tc,
2429 #ifdef CONFIG_RFS_ACCEL
2430         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2431 #endif
2432         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2433 };
2434
2435 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2436                         struct mlx4_en_port_profile *prof)
2437 {
2438         struct net_device *dev;
2439         struct mlx4_en_priv *priv;
2440         int i;
2441         int err;
2442         u64 mac_u64;
2443
2444         dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2445                                  MAX_TX_RINGS, MAX_RX_RINGS);
2446         if (dev == NULL)
2447                 return -ENOMEM;
2448
2449         netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2450         netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2451
2452         SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
2453         dev->dev_port = port - 1;
2454
2455         /*
2456          * Initialize driver private data
2457          */
2458
2459         priv = netdev_priv(dev);
2460         memset(priv, 0, sizeof(struct mlx4_en_priv));
2461         priv->dev = dev;
2462         priv->mdev = mdev;
2463         priv->ddev = &mdev->pdev->dev;
2464         priv->prof = prof;
2465         priv->port = port;
2466         priv->port_up = false;
2467         priv->flags = prof->flags;
2468         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2469                         MLX4_WQE_CTRL_SOLICITED);
2470         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2471         priv->tx_ring_num = prof->tx_ring_num;
2472         priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
2473
2474         priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2475                                 GFP_KERNEL);
2476         if (!priv->tx_ring) {
2477                 err = -ENOMEM;
2478                 goto out;
2479         }
2480         priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2481                               GFP_KERNEL);
2482         if (!priv->tx_cq) {
2483                 err = -ENOMEM;
2484                 goto out;
2485         }
2486         priv->rx_ring_num = prof->rx_ring_num;
2487         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2488         priv->mac_index = -1;
2489         priv->msg_enable = MLX4_EN_MSG_LEVEL;
2490         spin_lock_init(&priv->stats_lock);
2491         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2492         INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2493         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2494         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2495         INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2496 #ifdef CONFIG_MLX4_EN_VXLAN
2497         INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
2498         INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
2499 #endif
2500 #ifdef CONFIG_MLX4_EN_DCB
2501         if (!mlx4_is_slave(priv->mdev->dev)) {
2502                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
2503                         dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2504                 } else {
2505                         en_info(priv, "enabling only PFC DCB ops\n");
2506                         dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2507                 }
2508         }
2509 #endif
2510
2511         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2512                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2513
2514         /* Query for default mac and max mtu */
2515         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2516
2517         /* Set default MAC */
2518         dev->addr_len = ETH_ALEN;
2519         mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2520         if (!is_valid_ether_addr(dev->dev_addr)) {
2521                 if (mlx4_is_slave(priv->mdev->dev)) {
2522                         eth_hw_addr_random(dev);
2523                         en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
2524                         mac_u64 = mlx4_mac_to_u64(dev->dev_addr);
2525                         mdev->dev->caps.def_mac[priv->port] = mac_u64;
2526                 } else {
2527                         en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2528                                priv->port, dev->dev_addr);
2529                         err = -EINVAL;
2530                         goto out;
2531                 }
2532         }
2533
2534         memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
2535
2536         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2537                                           DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2538         err = mlx4_en_alloc_resources(priv);
2539         if (err)
2540                 goto out;
2541
2542 #ifdef CONFIG_RFS_ACCEL
2543         INIT_LIST_HEAD(&priv->filters);
2544         spin_lock_init(&priv->filters_lock);
2545 #endif
2546
2547         /* Initialize time stamping config */
2548         priv->hwtstamp_config.flags = 0;
2549         priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
2550         priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2551
2552         /* Allocate page for receive rings */
2553         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2554                                 MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2555         if (err) {
2556                 en_err(priv, "Failed to allocate page for rx qps\n");
2557                 goto out;
2558         }
2559         priv->allocated = 1;
2560
2561         /*
2562          * Initialize netdev entry points
2563          */
2564         if (mlx4_is_master(priv->mdev->dev))
2565                 dev->netdev_ops = &mlx4_netdev_ops_master;
2566         else
2567                 dev->netdev_ops = &mlx4_netdev_ops;
2568         dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2569         netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2570         netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2571
2572         dev->ethtool_ops = &mlx4_en_ethtool_ops;
2573
2574         /*
2575          * Set driver features
2576          */
2577         dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2578         if (mdev->LSO_support)
2579                 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2580
2581         dev->vlan_features = dev->hw_features;
2582
2583         dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2584         dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2585                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2586                         NETIF_F_HW_VLAN_CTAG_FILTER;
2587         dev->hw_features |= NETIF_F_LOOPBACK;
2588
2589         if (mdev->dev->caps.steering_mode ==
2590             MLX4_STEERING_MODE_DEVICE_MANAGED)
2591                 dev->hw_features |= NETIF_F_NTUPLE;
2592
2593         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
2594                 dev->priv_flags |= IFF_UNICAST_FLT;
2595
2596         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2597                 dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
2598                                         NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL;
2599                 dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
2600                 dev->features    |= NETIF_F_GSO_UDP_TUNNEL;
2601         }
2602
2603         mdev->pndev[port] = dev;
2604
2605         netif_carrier_off(dev);
2606         mlx4_en_set_default_moderation(priv);
2607
2608         err = register_netdev(dev);
2609         if (err) {
2610                 en_err(priv, "Netdev registration failed for port %d\n", port);
2611                 goto out;
2612         }
2613         priv->registered = 1;
2614
2615         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2616         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2617
2618         mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
2619
2620         /* Configure port */
2621         mlx4_en_calc_rx_buf(dev);
2622         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2623                                     priv->rx_skb_size + ETH_FCS_LEN,
2624                                     prof->tx_pause, prof->tx_ppp,
2625                                     prof->rx_pause, prof->rx_ppp);
2626         if (err) {
2627                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
2628                        priv->port, err);
2629                 goto out;
2630         }
2631
2632         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2633                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
2634                 if (err) {
2635                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
2636                                err);
2637                         goto out;
2638                 }
2639         }
2640
2641         /* Init port */
2642         en_warn(priv, "Initializing port\n");
2643         err = mlx4_INIT_PORT(mdev->dev, priv->port);
2644         if (err) {
2645                 en_err(priv, "Failed Initializing port\n");
2646                 goto out;
2647         }
2648         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2649
2650         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2651                 queue_delayed_work(mdev->workqueue, &priv->service_task,
2652                                    SERVICE_TASK_DELAY);
2653
2654         return 0;
2655
2656 out:
2657         mlx4_en_destroy_netdev(dev);
2658         return err;
2659 }
2660