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
 478         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
 479                 return 0; /* do nothing */
 480
 481         err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
 482                                     MLX4_DOMAIN_NIC, reg_id);
 483         if (err) {
 484                 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
 485                 return err;
 486         }
 487         en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
 488         return 0;
 489 }
 490
 491
 492 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
 493                                 unsigned char *mac, int *qpn, u64 *reg_id)
 494 {
 495         struct mlx4_en_dev *mdev = priv->mdev;
 496         struct mlx4_dev *dev = mdev->dev;
 497         int err;
 498
 499         switch (dev->caps.steering_mode) {
 500         case MLX4_STEERING_MODE_B0: {
 501                 struct mlx4_qp qp;
 502                 u8 gid[16] = {0};
 503
 504                 qp.qpn = *qpn;
 505                 memcpy(&gid[10], mac, ETH_ALEN);
 506                 gid[5] = priv->port;
 507
 508                 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
 509                 break;
 510         }
 511         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 512                 struct mlx4_spec_list spec_eth = { {NULL} };
 513                 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 514
 515                 struct mlx4_net_trans_rule rule = {
 516                         .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 517                         .exclusive = 0,
 518                         .allow_loopback = 1,
 519                         .promisc_mode = MLX4_FS_REGULAR,
 520                         .priority = MLX4_DOMAIN_NIC,
 521                 };
 522
 523                 rule.port = priv->port;
 524                 rule.qpn = *qpn;
 525                 INIT_LIST_HEAD(&rule.list);
 526
 527                 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
 528                 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
 529                 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 530                 list_add_tail(&spec_eth.list, &rule.list);
 531
 532                 err = mlx4_flow_attach(dev, &rule, reg_id);
 533                 break;
 534         }
 535         default:
 536                 return -EINVAL;
 537         }
 538         if (err)
 539                 en_warn(priv, "Failed Attaching Unicast\n");
 540
 541         return err;
 542 }
 543
 544 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
 545                                      unsigned char *mac, int qpn, u64 reg_id)
 546 {
 547         struct mlx4_en_dev *mdev = priv->mdev;
 548         struct mlx4_dev *dev = mdev->dev;
 549
 550         switch (dev->caps.steering_mode) {
 551         case MLX4_STEERING_MODE_B0: {
 552                 struct mlx4_qp qp;
 553                 u8 gid[16] = {0};
 554
 555                 qp.qpn = qpn;
 556                 memcpy(&gid[10], mac, ETH_ALEN);
 557                 gid[5] = priv->port;
 558
 559                 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
 560                 break;
 561         }
 562         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 563                 mlx4_flow_detach(dev, reg_id);
 564                 break;
 565         }
 566         default:
 567                 en_err(priv, "Invalid steering mode.\n");
 568         }
 569 }
 570
 571 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
 572 {
 573         struct mlx4_en_dev *mdev = priv->mdev;
 574         struct mlx4_dev *dev = mdev->dev;
 575         struct mlx4_mac_entry *entry;
 576         int index = 0;
 577         int err = 0;
 578         u64 reg_id = 0;
 579         int *qpn = &priv->base_qpn;
 580         u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 581
 582         en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
 583                priv->dev->dev_addr);
 584         index = mlx4_register_mac(dev, priv->port, mac);
 585         if (index < 0) {
 586                 err = index;
 587                 en_err(priv, "Failed adding MAC: %pM\n",
 588                        priv->dev->dev_addr);
 589                 return err;
 590         }
 591
 592         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 593                 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
 594                 *qpn = base_qpn + index;
 595                 return 0;
 596         }
 597
 598         err = mlx4_qp_reserve_range(dev, 1, 1, qpn);
 599         en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
 600         if (err) {
 601                 en_err(priv, "Failed to reserve qp for mac registration\n");
 602                 goto qp_err;
 603         }
 604
 605         err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
 606         if (err)
 607                 goto steer_err;
 608
 609         err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
 610                                        &priv->tunnel_reg_id);
 611         if (err)
 612                 goto tunnel_err;
 613
 614         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 615         if (!entry) {
 616                 err = -ENOMEM;
 617                 goto alloc_err;
 618         }
 619         memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
 620         memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
 621         entry->reg_id = reg_id;
 622
 623         hlist_add_head_rcu(&entry->hlist,
 624                            &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
 625
 626         return 0;
 627
 628 alloc_err:
 629         if (priv->tunnel_reg_id)
 630                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 631 tunnel_err:
 632         mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
 633
 634 steer_err:
 635         mlx4_qp_release_range(dev, *qpn, 1);
 636
 637 qp_err:
 638         mlx4_unregister_mac(dev, priv->port, mac);
 639         return err;
 640 }
 641
 642 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
 643 {
 644         struct mlx4_en_dev *mdev = priv->mdev;
 645         struct mlx4_dev *dev = mdev->dev;
 646         int qpn = priv->base_qpn;
 647         u64 mac;
 648
 649         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 650                 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 651                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 652                        priv->dev->dev_addr);
 653                 mlx4_unregister_mac(dev, priv->port, mac);
 654         } else {
 655                 struct mlx4_mac_entry *entry;
 656                 struct hlist_node *tmp;
 657                 struct hlist_head *bucket;
 658                 unsigned int i;
 659
 660                 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
 661                         bucket = &priv->mac_hash[i];
 662                         hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 663                                 mac = mlx4_mac_to_u64(entry->mac);
 664                                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 665                                        entry->mac);
 666                                 mlx4_en_uc_steer_release(priv, entry->mac,
 667                                                          qpn, entry->reg_id);
 668
 669                                 mlx4_unregister_mac(dev, priv->port, mac);
 670                                 hlist_del_rcu(&entry->hlist);
 671                                 kfree_rcu(entry, rcu);
 672                         }
 673                 }
 674
 675                 if (priv->tunnel_reg_id) {
 676                         mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 677                         priv->tunnel_reg_id = 0;
 678                 }
 679
 680                 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
 681                        priv->port, qpn);
 682                 mlx4_qp_release_range(dev, qpn, 1);
 683                 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
 684         }
 685 }
 686
 687 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
 688                                unsigned char *new_mac, unsigned char *prev_mac)
 689 {
 690         struct mlx4_en_dev *mdev = priv->mdev;
 691         struct mlx4_dev *dev = mdev->dev;
 692         int err = 0;
 693         u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
 694
 695         if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
 696                 struct hlist_head *bucket;
 697                 unsigned int mac_hash;
 698                 struct mlx4_mac_entry *entry;
 699                 struct hlist_node *tmp;
 700                 u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
 701
 702                 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
 703                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 704                         if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
 705                                 mlx4_en_uc_steer_release(priv, entry->mac,
 706                                                          qpn, entry->reg_id);
 707                                 mlx4_unregister_mac(dev, priv->port,
 708                                                     prev_mac_u64);
 709                                 hlist_del_rcu(&entry->hlist);
 710                                 synchronize_rcu();
 711                                 memcpy(entry->mac, new_mac, ETH_ALEN);
 712                                 entry->reg_id = 0;
 713                                 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
 714                                 hlist_add_head_rcu(&entry->hlist,
 715                                                    &priv->mac_hash[mac_hash]);
 716                                 mlx4_register_mac(dev, priv->port, new_mac_u64);
 717                                 err = mlx4_en_uc_steer_add(priv, new_mac,
 718                                                            &qpn,
 719                                                            &entry->reg_id);
 720                                 if (err)
 721                                         return err;
 722                                 if (priv->tunnel_reg_id) {
 723                                         mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 724                                         priv->tunnel_reg_id = 0;
 725                                 }
 726                                 err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
 727                                                                &priv->tunnel_reg_id);
 728                                 return err;
 729                         }
 730                 }
 731                 return -EINVAL;
 732         }
 733
 734         return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
 735 }
 736
 737 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
 738                               unsigned char new_mac[ETH_ALEN + 2])
 739 {
 740         int err = 0;
 741
 742         if (priv->port_up) {
 743                 /* Remove old MAC and insert the new one */
 744                 err = mlx4_en_replace_mac(priv, priv->base_qpn,
 745                                           new_mac, priv->current_mac);
 746                 if (err)
 747                         en_err(priv, "Failed changing HW MAC address\n");
 748         } else
 749                 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
 750
 751         if (!err)
 752                 memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
 753
 754         return err;
 755 }
 756
 757 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
 758 {
 759         struct mlx4_en_priv *priv = netdev_priv(dev);
 760         struct mlx4_en_dev *mdev = priv->mdev;
 761         struct sockaddr *saddr = addr;
 762         unsigned char new_mac[ETH_ALEN + 2];
 763         int err;
 764
 765         if (!is_valid_ether_addr(saddr->sa_data))
 766                 return -EADDRNOTAVAIL;
 767
 768         mutex_lock(&mdev->state_lock);
 769         memcpy(new_mac, saddr->sa_data, ETH_ALEN);
 770         err = mlx4_en_do_set_mac(priv, new_mac);
 771         if (!err)
 772                 memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
 773         mutex_unlock(&mdev->state_lock);
 774
 775         return err;
 776 }
 777
 778 static void mlx4_en_clear_list(struct net_device *dev)
 779 {
 780         struct mlx4_en_priv *priv = netdev_priv(dev);
 781         struct mlx4_en_mc_list *tmp, *mc_to_del;
 782
 783         list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
 784                 list_del(&mc_to_del->list);
 785                 kfree(mc_to_del);
 786         }
 787 }
 788
 789 static void mlx4_en_cache_mclist(struct net_device *dev)
 790 {
 791         struct mlx4_en_priv *priv = netdev_priv(dev);
 792         struct netdev_hw_addr *ha;
 793         struct mlx4_en_mc_list *tmp;
 794
 795         mlx4_en_clear_list(dev);
 796         netdev_for_each_mc_addr(ha, dev) {
 797                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 798                 if (!tmp) {
 799                         mlx4_en_clear_list(dev);
 800                         return;
 801                 }
 802                 memcpy(tmp->addr, ha->addr, ETH_ALEN);
 803                 list_add_tail(&tmp->list, &priv->mc_list);
 804         }
 805 }
 806
 807 static void update_mclist_flags(struct mlx4_en_priv *priv,
 808                                 struct list_head *dst,
 809                                 struct list_head *src)
 810 {
 811         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 812         bool found;
 813
 814         /* Find all the entries that should be removed from dst,
 815          * These are the entries that are not found in src
 816          */
 817         list_for_each_entry(dst_tmp, dst, list) {
 818                 found = false;
 819                 list_for_each_entry(src_tmp, src, list) {
 820                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 821                                 found = true;
 822                                 break;
 823                         }
 824                 }
 825                 if (!found)
 826                         dst_tmp->action = MCLIST_REM;
 827         }
 828
 829         /* Add entries that exist in src but not in dst
 830          * mark them as need to add
 831          */
 832         list_for_each_entry(src_tmp, src, list) {
 833                 found = false;
 834                 list_for_each_entry(dst_tmp, dst, list) {
 835                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 836                                 dst_tmp->action = MCLIST_NONE;
 837                                 found = true;
 838                                 break;
 839                         }
 840                 }
 841                 if (!found) {
 842                         new_mc = kmemdup(src_tmp,
 843                                          sizeof(struct mlx4_en_mc_list),
 844                                          GFP_KERNEL);
 845                         if (!new_mc)
 846                                 return;
 847
 848                         new_mc->action = MCLIST_ADD;
 849                         list_add_tail(&new_mc->list, dst);
 850                 }
 851         }
 852 }
 853
 854 static void mlx4_en_set_rx_mode(struct net_device *dev)
 855 {
 856         struct mlx4_en_priv *priv = netdev_priv(dev);
 857
 858         if (!priv->port_up)
 859                 return;
 860
 861         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 862 }
 863
 864 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 865                                      struct mlx4_en_dev *mdev)
 866 {
 867         int err = 0;
 868
 869         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 870                 if (netif_msg_rx_status(priv))
 871                         en_warn(priv, "Entering promiscuous mode\n");
 872                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 873
 874                 /* Enable promiscouos mode */
 875                 switch (mdev->dev->caps.steering_mode) {
 876                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 877                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 878                                                           priv->port,
 879                                                           priv->base_qpn,
 880                                                           MLX4_FS_ALL_DEFAULT);
 881                         if (err)
 882                                 en_err(priv, "Failed enabling promiscuous mode\n");
 883                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 884                         break;
 885
 886                 case MLX4_STEERING_MODE_B0:
 887                         err = mlx4_unicast_promisc_add(mdev->dev,
 888                                                        priv->base_qpn,
 889                                                        priv->port);
 890                         if (err)
 891                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 892
 893                         /* Add the default qp number as multicast
 894                          * promisc
 895                          */
 896                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 897                                 err = mlx4_multicast_promisc_add(mdev->dev,
 898                                                                  priv->base_qpn,
 899                                                                  priv->port);
 900                                 if (err)
 901                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 902                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 903                         }
 904                         break;
 905
 906                 case MLX4_STEERING_MODE_A0:
 907                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 908                                                      priv->port,
 909                                                      priv->base_qpn,
 910                                                      1);
 911                         if (err)
 912                                 en_err(priv, "Failed enabling promiscuous mode\n");
 913                         break;
 914                 }
 915
 916                 /* Disable port multicast filter (unconditionally) */
 917                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 918                                           0, MLX4_MCAST_DISABLE);
 919                 if (err)
 920                         en_err(priv, "Failed disabling multicast filter\n");
 921         }
 922 }
 923
 924 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 925                                        struct mlx4_en_dev *mdev)
 926 {
 927         int err = 0;
 928
 929         if (netif_msg_rx_status(priv))
 930                 en_warn(priv, "Leaving promiscuous mode\n");
 931         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 932
 933         /* Disable promiscouos mode */
 934         switch (mdev->dev->caps.steering_mode) {
 935         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 936                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 937                                                      priv->port,
 938                                                      MLX4_FS_ALL_DEFAULT);
 939                 if (err)
 940                         en_err(priv, "Failed disabling promiscuous mode\n");
 941                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 942                 break;
 943
 944         case MLX4_STEERING_MODE_B0:
 945                 err = mlx4_unicast_promisc_remove(mdev->dev,
 946                                                   priv->base_qpn,
 947                                                   priv->port);
 948                 if (err)
 949                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 950                 /* Disable Multicast promisc */
 951                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 952                         err = mlx4_multicast_promisc_remove(mdev->dev,
 953                                                             priv->base_qpn,
 954                                                             priv->port);
 955                         if (err)
 956                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 957                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 958                 }
 959                 break;
 960
 961         case MLX4_STEERING_MODE_A0:
 962                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 963                                              priv->port,
 964                                              priv->base_qpn, 0);
 965                 if (err)
 966                         en_err(priv, "Failed disabling promiscuous mode\n");
 967                 break;
 968         }
 969 }
 970
 971 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
 972                                  struct net_device *dev,
 973                                  struct mlx4_en_dev *mdev)
 974 {
 975         struct mlx4_en_mc_list *mclist, *tmp;
 976         u64 mcast_addr = 0;
 977         u8 mc_list[16] = {0};
 978         int err = 0;
 979
 980         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
 981         if (dev->flags & IFF_ALLMULTI) {
 982                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 983                                           0, MLX4_MCAST_DISABLE);
 984                 if (err)
 985                         en_err(priv, "Failed disabling multicast filter\n");
 986
 987                 /* Add the default qp number as multicast promisc */
 988                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 989                         switch (mdev->dev->caps.steering_mode) {
 990                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 991                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
 992                                                                   priv->port,
 993                                                                   priv->base_qpn,
 994                                                                   MLX4_FS_MC_DEFAULT);
 995                                 break;
 996
 997                         case MLX4_STEERING_MODE_B0:
 998                                 err = mlx4_multicast_promisc_add(mdev->dev,
 999                                                                  priv->base_qpn,
1000                                                                  priv->port);
1001                                 break;
1002
1003                         case MLX4_STEERING_MODE_A0:
1004                                 break;
1005                         }
1006                         if (err)
1007                                 en_err(priv, "Failed entering multicast promisc mode\n");
1008                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1009                 }
1010         } else {
1011                 /* Disable Multicast promisc */
1012                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1013                         switch (mdev->dev->caps.steering_mode) {
1014                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
1015                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
1016                                                                      priv->port,
1017                                                                      MLX4_FS_MC_DEFAULT);
1018                                 break;
1019
1020                         case MLX4_STEERING_MODE_B0:
1021                                 err = mlx4_multicast_promisc_remove(mdev->dev,
1022                                                                     priv->base_qpn,
1023                                                                     priv->port);
1024                                 break;
1025
1026                         case MLX4_STEERING_MODE_A0:
1027                                 break;
1028                         }
1029                         if (err)
1030                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
1031                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1032                 }
1033
1034                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1035                                           0, MLX4_MCAST_DISABLE);
1036                 if (err)
1037                         en_err(priv, "Failed disabling multicast filter\n");
1038
1039                 /* Flush mcast filter and init it with broadcast address */
1040                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1041                                     1, MLX4_MCAST_CONFIG);
1042
1043                 /* Update multicast list - we cache all addresses so they won't
1044                  * change while HW is updated holding the command semaphor */
1045                 netif_addr_lock_bh(dev);
1046                 mlx4_en_cache_mclist(dev);
1047                 netif_addr_unlock_bh(dev);
1048                 list_for_each_entry(mclist, &priv->mc_list, list) {
1049                         mcast_addr = mlx4_mac_to_u64(mclist->addr);
1050                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1051                                             mcast_addr, 0, MLX4_MCAST_CONFIG);
1052                 }
1053                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1054                                           0, MLX4_MCAST_ENABLE);
1055                 if (err)
1056                         en_err(priv, "Failed enabling multicast filter\n");
1057
1058                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1059                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1060                         if (mclist->action == MCLIST_REM) {
1061                                 /* detach this address and delete from list */
1062                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1063                                 mc_list[5] = priv->port;
1064                                 err = mlx4_multicast_detach(mdev->dev,
1065                                                             &priv->rss_map.indir_qp,
1066                                                             mc_list,
1067                                                             MLX4_PROT_ETH,
1068                                                             mclist->reg_id);
1069                                 if (err)
1070                                         en_err(priv, "Fail to detach multicast address\n");
1071
1072                                 if (mclist->tunnel_reg_id) {
1073                                         err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1074                                         if (err)
1075                                                 en_err(priv, "Failed to detach multicast address\n");
1076                                 }
1077
1078                                 /* remove from list */
1079                                 list_del(&mclist->list);
1080                                 kfree(mclist);
1081                         } else if (mclist->action == MCLIST_ADD) {
1082                                 /* attach the address */
1083                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1084                                 /* needed for B0 steering support */
1085                                 mc_list[5] = priv->port;
1086                                 err = mlx4_multicast_attach(mdev->dev,
1087                                                             &priv->rss_map.indir_qp,
1088                                                             mc_list,
1089                                                             priv->port, 0,
1090                                                             MLX4_PROT_ETH,
1091                                                             &mclist->reg_id);
1092                                 if (err)
1093                                         en_err(priv, "Fail to attach multicast address\n");
1094
1095                                 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1096                                                                &mclist->tunnel_reg_id);
1097                                 if (err)
1098                                         en_err(priv, "Failed to attach multicast address\n");
1099                         }
1100                 }
1101         }
1102 }
1103
1104 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1105                                  struct net_device *dev,
1106                                  struct mlx4_en_dev *mdev)
1107 {
1108         struct netdev_hw_addr *ha;
1109         struct mlx4_mac_entry *entry;
1110         struct hlist_node *tmp;
1111         bool found;
1112         u64 mac;
1113         int err = 0;
1114         struct hlist_head *bucket;
1115         unsigned int i;
1116         int removed = 0;
1117         u32 prev_flags;
1118
1119         /* Note that we do not need to protect our mac_hash traversal with rcu,
1120          * since all modification code is protected by mdev->state_lock
1121          */
1122
1123         /* find what to remove */
1124         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1125                 bucket = &priv->mac_hash[i];
1126                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1127                         found = false;
1128                         netdev_for_each_uc_addr(ha, dev) {
1129                                 if (ether_addr_equal_64bits(entry->mac,
1130                                                             ha->addr)) {
1131                                         found = true;
1132                                         break;
1133                                 }
1134                         }
1135
1136                         /* MAC address of the port is not in uc list */
1137                         if (ether_addr_equal_64bits(entry->mac,
1138                                                     priv->current_mac))
1139                                 found = true;
1140
1141                         if (!found) {
1142                                 mac = mlx4_mac_to_u64(entry->mac);
1143                                 mlx4_en_uc_steer_release(priv, entry->mac,
1144                                                          priv->base_qpn,
1145                                                          entry->reg_id);
1146                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1147
1148                                 hlist_del_rcu(&entry->hlist);
1149                                 kfree_rcu(entry, rcu);
1150                                 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1151                                        entry->mac, priv->port);
1152                                 ++removed;
1153                         }
1154                 }
1155         }
1156
1157         /* if we didn't remove anything, there is no use in trying to add
1158          * again once we are in a forced promisc mode state
1159          */
1160         if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1161                 return;
1162
1163         prev_flags = priv->flags;
1164         priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1165
1166         /* find what to add */
1167         netdev_for_each_uc_addr(ha, dev) {
1168                 found = false;
1169                 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1170                 hlist_for_each_entry(entry, bucket, hlist) {
1171                         if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1172                                 found = true;
1173                                 break;
1174                         }
1175                 }
1176
1177                 if (!found) {
1178                         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1179                         if (!entry) {
1180                                 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1181                                        ha->addr, priv->port);
1182                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1183                                 break;
1184                         }
1185                         mac = mlx4_mac_to_u64(ha->addr);
1186                         memcpy(entry->mac, ha->addr, ETH_ALEN);
1187                         err = mlx4_register_mac(mdev->dev, priv->port, mac);
1188                         if (err < 0) {
1189                                 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1190                                        ha->addr, priv->port, err);
1191                                 kfree(entry);
1192                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1193                                 break;
1194                         }
1195                         err = mlx4_en_uc_steer_add(priv, ha->addr,
1196                                                    &priv->base_qpn,
1197                                                    &entry->reg_id);
1198                         if (err) {
1199                                 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1200                                        ha->addr, priv->port, err);
1201                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1202                                 kfree(entry);
1203                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1204                                 break;
1205                         } else {
1206                                 unsigned int mac_hash;
1207                                 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1208                                        ha->addr, priv->port);
1209                                 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1210                                 bucket = &priv->mac_hash[mac_hash];
1211                                 hlist_add_head_rcu(&entry->hlist, bucket);
1212                         }
1213                 }
1214         }
1215
1216         if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1217                 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1218                         priv->port);
1219         } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1220                 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1221                         priv->port);
1222         }
1223 }
1224
1225 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1226 {
1227         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1228                                                  rx_mode_task);
1229         struct mlx4_en_dev *mdev = priv->mdev;
1230         struct net_device *dev = priv->dev;
1231
1232         mutex_lock(&mdev->state_lock);
1233         if (!mdev->device_up) {
1234                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1235                 goto out;
1236         }
1237         if (!priv->port_up) {
1238                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1239                 goto out;
1240         }
1241
1242         if (!netif_carrier_ok(dev)) {
1243                 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1244                         if (priv->port_state.link_state) {
1245                                 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1246                                 netif_carrier_on(dev);
1247                                 en_dbg(LINK, priv, "Link Up\n");
1248                         }
1249                 }
1250         }
1251
1252         if (dev->priv_flags & IFF_UNICAST_FLT)
1253                 mlx4_en_do_uc_filter(priv, dev, mdev);
1254
1255         /* Promsicuous mode: disable all filters */
1256         if ((dev->flags & IFF_PROMISC) ||
1257             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1258                 mlx4_en_set_promisc_mode(priv, mdev);
1259                 goto out;
1260         }
1261
1262         /* Not in promiscuous mode */
1263         if (priv->flags & MLX4_EN_FLAG_PROMISC)
1264                 mlx4_en_clear_promisc_mode(priv, mdev);
1265
1266         mlx4_en_do_multicast(priv, dev, mdev);
1267 out:
1268         mutex_unlock(&mdev->state_lock);
1269 }
1270
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 static void mlx4_en_netpoll(struct net_device *dev)
1273 {
1274         struct mlx4_en_priv *priv = netdev_priv(dev);
1275         struct mlx4_en_cq *cq;
1276         int i;
1277
1278         for (i = 0; i < priv->rx_ring_num; i++) {
1279                 cq = priv->rx_cq[i];
1280                 napi_schedule(&cq->napi);
1281         }
1282 }
1283 #endif
1284
1285 static void mlx4_en_tx_timeout(struct net_device *dev)
1286 {
1287         struct mlx4_en_priv *priv = netdev_priv(dev);
1288         struct mlx4_en_dev *mdev = priv->mdev;
1289         int i;
1290
1291         if (netif_msg_timer(priv))
1292                 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1293
1294         for (i = 0; i < priv->tx_ring_num; i++) {
1295                 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1296                         continue;
1297                 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1298                         i, priv->tx_ring[i]->qpn, priv->tx_ring[i]->cqn,
1299                         priv->tx_ring[i]->cons, priv->tx_ring[i]->prod);
1300         }
1301
1302         priv->port_stats.tx_timeout++;
1303         en_dbg(DRV, priv, "Scheduling watchdog\n");
1304         queue_work(mdev->workqueue, &priv->watchdog_task);
1305 }
1306
1307
1308 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
1309 {
1310         struct mlx4_en_priv *priv = netdev_priv(dev);
1311
1312         spin_lock_bh(&priv->stats_lock);
1313         memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
1314         spin_unlock_bh(&priv->stats_lock);
1315
1316         return &priv->ret_stats;
1317 }
1318
1319 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1320 {
1321         struct mlx4_en_cq *cq;
1322         int i;
1323
1324         /* If we haven't received a specific coalescing setting
1325          * (module param), we set the moderation parameters as follows:
1326          * - moder_cnt is set to the number of mtu sized packets to
1327          *   satisfy our coalescing target.
1328          * - moder_time is set to a fixed value.
1329          */
1330         priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1331         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1332         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1333         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1334         en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1335                priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1336
1337         /* Setup cq moderation params */
1338         for (i = 0; i < priv->rx_ring_num; i++) {
1339                 cq = priv->rx_cq[i];
1340                 cq->moder_cnt = priv->rx_frames;
1341                 cq->moder_time = priv->rx_usecs;
1342                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1343                 priv->last_moder_packets[i] = 0;
1344                 priv->last_moder_bytes[i] = 0;
1345         }
1346
1347         for (i = 0; i < priv->tx_ring_num; i++) {
1348                 cq = priv->tx_cq[i];
1349                 cq->moder_cnt = priv->tx_frames;
1350                 cq->moder_time = priv->tx_usecs;
1351         }
1352
1353         /* Reset auto-moderation params */
1354         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1355         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1356         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1357         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1358         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1359         priv->adaptive_rx_coal = 1;
1360         priv->last_moder_jiffies = 0;
1361         priv->last_moder_tx_packets = 0;
1362 }
1363
1364 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1365 {
1366         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1367         struct mlx4_en_cq *cq;
1368         unsigned long packets;
1369         unsigned long rate;
1370         unsigned long avg_pkt_size;
1371         unsigned long rx_packets;
1372         unsigned long rx_bytes;
1373         unsigned long rx_pkt_diff;
1374         int moder_time;
1375         int ring, err;
1376
1377         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1378                 return;
1379
1380         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1381                 spin_lock_bh(&priv->stats_lock);
1382                 rx_packets = priv->rx_ring[ring]->packets;
1383                 rx_bytes = priv->rx_ring[ring]->bytes;
1384                 spin_unlock_bh(&priv->stats_lock);
1385
1386                 rx_pkt_diff = ((unsigned long) (rx_packets -
1387                                 priv->last_moder_packets[ring]));
1388                 packets = rx_pkt_diff;
1389                 rate = packets * HZ / period;
1390                 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1391                                 priv->last_moder_bytes[ring])) / packets : 0;
1392
1393                 /* Apply auto-moderation only when packet rate
1394                  * exceeds a rate that it matters */
1395                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1396                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1397                         if (rate < priv->pkt_rate_low)
1398                                 moder_time = priv->rx_usecs_low;
1399                         else if (rate > priv->pkt_rate_high)
1400                                 moder_time = priv->rx_usecs_high;
1401                         else
1402                                 moder_time = (rate - priv->pkt_rate_low) *
1403                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1404                                         (priv->pkt_rate_high - priv->pkt_rate_low) +
1405                                         priv->rx_usecs_low;
1406                 } else {
1407                         moder_time = priv->rx_usecs_low;
1408                 }
1409
1410                 if (moder_time != priv->last_moder_time[ring]) {
1411                         priv->last_moder_time[ring] = moder_time;
1412                         cq = priv->rx_cq[ring];
1413                         cq->moder_time = moder_time;
1414                         cq->moder_cnt = priv->rx_frames;
1415                         err = mlx4_en_set_cq_moder(priv, cq);
1416                         if (err)
1417                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1418                                        ring);
1419                 }
1420                 priv->last_moder_packets[ring] = rx_packets;
1421                 priv->last_moder_bytes[ring] = rx_bytes;
1422         }
1423
1424         priv->last_moder_jiffies = jiffies;
1425 }
1426
1427 static void mlx4_en_do_get_stats(struct work_struct *work)
1428 {
1429         struct delayed_work *delay = to_delayed_work(work);
1430         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1431                                                  stats_task);
1432         struct mlx4_en_dev *mdev = priv->mdev;
1433         int err;
1434
1435         mutex_lock(&mdev->state_lock);
1436         if (mdev->device_up) {
1437                 if (priv->port_up) {
1438                         err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1439                         if (err)
1440                                 en_dbg(HW, priv, "Could not update stats\n");
1441
1442                         mlx4_en_auto_moderation(priv);
1443                 }
1444
1445                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1446         }
1447         if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1448                 mlx4_en_do_set_mac(priv, priv->current_mac);
1449                 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1450         }
1451         mutex_unlock(&mdev->state_lock);
1452 }
1453
1454 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1455  * periodically
1456  */
1457 static void mlx4_en_service_task(struct work_struct *work)
1458 {
1459         struct delayed_work *delay = to_delayed_work(work);
1460         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1461                                                  service_task);
1462         struct mlx4_en_dev *mdev = priv->mdev;
1463
1464         mutex_lock(&mdev->state_lock);
1465         if (mdev->device_up) {
1466                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1467                         mlx4_en_ptp_overflow_check(mdev);
1468
1469                 queue_delayed_work(mdev->workqueue, &priv->service_task,
1470                                    SERVICE_TASK_DELAY);
1471         }
1472         mutex_unlock(&mdev->state_lock);
1473 }
1474
1475 static void mlx4_en_linkstate(struct work_struct *work)
1476 {
1477         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1478                                                  linkstate_task);
1479         struct mlx4_en_dev *mdev = priv->mdev;
1480         int linkstate = priv->link_state;
1481
1482         mutex_lock(&mdev->state_lock);
1483         /* If observable port state changed set carrier state and
1484          * report to system log */
1485         if (priv->last_link_state != linkstate) {
1486                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1487                         en_info(priv, "Link Down\n");
1488                         netif_carrier_off(priv->dev);
1489                 } else {
1490                         en_info(priv, "Link Up\n");
1491                         netif_carrier_on(priv->dev);
1492                 }
1493         }
1494         priv->last_link_state = linkstate;
1495         mutex_unlock(&mdev->state_lock);
1496 }
1497
1498 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1499 {
1500         struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1501         int numa_node = priv->mdev->dev->numa_node;
1502         int ret = 0;
1503
1504         if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1505                 return -ENOMEM;
1506
1507         ret = cpumask_set_cpu_local_first(ring_idx, numa_node,
1508                                           ring->affinity_mask);
1509         if (ret)
1510                 free_cpumask_var(ring->affinity_mask);
1511
1512         return ret;
1513 }
1514
1515 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1516 {
1517         free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1518 }
1519
1520 int mlx4_en_start_port(struct net_device *dev)
1521 {
1522         struct mlx4_en_priv *priv = netdev_priv(dev);
1523         struct mlx4_en_dev *mdev = priv->mdev;
1524         struct mlx4_en_cq *cq;
1525         struct mlx4_en_tx_ring *tx_ring;
1526         int rx_index = 0;
1527         int tx_index = 0;
1528         int err = 0;
1529         int i;
1530         int j;
1531         u8 mc_list[16] = {0};
1532
1533         if (priv->port_up) {
1534                 en_dbg(DRV, priv, "start port called while port already up\n");
1535                 return 0;
1536         }
1537
1538         INIT_LIST_HEAD(&priv->mc_list);
1539         INIT_LIST_HEAD(&priv->curr_list);
1540         INIT_LIST_HEAD(&priv->ethtool_list);
1541         memset(&priv->ethtool_rules[0], 0,
1542                sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1543
1544         /* Calculate Rx buf size */
1545         dev->mtu = min(dev->mtu, priv->max_mtu);
1546         mlx4_en_calc_rx_buf(dev);
1547         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1548
1549         /* Configure rx cq's and rings */
1550         err = mlx4_en_activate_rx_rings(priv);
1551         if (err) {
1552                 en_err(priv, "Failed to activate RX rings\n");
1553                 return err;
1554         }
1555         for (i = 0; i < priv->rx_ring_num; i++) {
1556                 cq = priv->rx_cq[i];
1557
1558                 mlx4_en_cq_init_lock(cq);
1559
1560                 err = mlx4_en_init_affinity_hint(priv, i);
1561                 if (err) {
1562                         en_err(priv, "Failed preparing IRQ affinity hint\n");
1563                         goto cq_err;
1564                 }
1565
1566                 err = mlx4_en_activate_cq(priv, cq, i);
1567                 if (err) {
1568                         en_err(priv, "Failed activating Rx CQ\n");
1569                         mlx4_en_free_affinity_hint(priv, i);
1570                         goto cq_err;
1571                 }
1572                 for (j = 0; j < cq->size; j++)
1573                         cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1574                 err = mlx4_en_set_cq_moder(priv, cq);
1575                 if (err) {
1576                         en_err(priv, "Failed setting cq moderation parameters\n");
1577                         mlx4_en_deactivate_cq(priv, cq);
1578                         mlx4_en_free_affinity_hint(priv, i);
1579                         goto cq_err;
1580                 }
1581                 mlx4_en_arm_cq(priv, cq);
1582                 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1583                 ++rx_index;
1584         }
1585
1586         /* Set qp number */
1587         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1588         err = mlx4_en_get_qp(priv);
1589         if (err) {
1590                 en_err(priv, "Failed getting eth qp\n");
1591                 goto cq_err;
1592         }
1593         mdev->mac_removed[priv->port] = 0;
1594
1595         err = mlx4_en_config_rss_steer(priv);
1596         if (err) {
1597                 en_err(priv, "Failed configuring rss steering\n");
1598                 goto mac_err;
1599         }
1600
1601         err = mlx4_en_create_drop_qp(priv);
1602         if (err)
1603                 goto rss_err;
1604
1605         /* Configure tx cq's and rings */
1606         for (i = 0; i < priv->tx_ring_num; i++) {
1607                 /* Configure cq */
1608                 cq = priv->tx_cq[i];
1609                 err = mlx4_en_activate_cq(priv, cq, i);
1610                 if (err) {
1611                         en_err(priv, "Failed allocating Tx CQ\n");
1612                         goto tx_err;
1613                 }
1614                 err = mlx4_en_set_cq_moder(priv, cq);
1615                 if (err) {
1616                         en_err(priv, "Failed setting cq moderation parameters\n");
1617                         mlx4_en_deactivate_cq(priv, cq);
1618                         goto tx_err;
1619                 }
1620                 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1621                 cq->buf->wqe_index = cpu_to_be16(0xffff);
1622
1623                 /* Configure ring */
1624                 tx_ring = priv->tx_ring[i];
1625                 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1626                         i / priv->num_tx_rings_p_up);
1627                 if (err) {
1628                         en_err(priv, "Failed allocating Tx ring\n");
1629                         mlx4_en_deactivate_cq(priv, cq);
1630                         goto tx_err;
1631                 }
1632                 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1633
1634                 /* Arm CQ for TX completions */
1635                 mlx4_en_arm_cq(priv, cq);
1636
1637                 /* Set initial ownership of all Tx TXBBs to SW (1) */
1638                 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1639                         *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1640                 ++tx_index;
1641         }
1642
1643         /* Configure port */
1644         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1645                                     priv->rx_skb_size + ETH_FCS_LEN,
1646                                     priv->prof->tx_pause,
1647                                     priv->prof->tx_ppp,
1648                                     priv->prof->rx_pause,
1649                                     priv->prof->rx_ppp);
1650         if (err) {
1651                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1652                        priv->port, err);
1653                 goto tx_err;
1654         }
1655         /* Set default qp number */
1656         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1657         if (err) {
1658                 en_err(priv, "Failed setting default qp numbers\n");
1659                 goto tx_err;
1660         }
1661
1662         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1663                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1664                 if (err) {
1665                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1666                                err);
1667                         goto tx_err;
1668                 }
1669         }
1670
1671         /* Init port */
1672         en_dbg(HW, priv, "Initializing port\n");
1673         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1674         if (err) {
1675                 en_err(priv, "Failed Initializing port\n");
1676                 goto tx_err;
1677         }
1678
1679         /* Attach rx QP to bradcast address */
1680         memset(&mc_list[10], 0xff, ETH_ALEN);
1681         mc_list[5] = priv->port; /* needed for B0 steering support */
1682         if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1683                                   priv->port, 0, MLX4_PROT_ETH,
1684                                   &priv->broadcast_id))
1685                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1686
1687         /* Must redo promiscuous mode setup. */
1688         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1689
1690         /* Schedule multicast task to populate multicast list */
1691         queue_work(mdev->workqueue, &priv->rx_mode_task);
1692
1693         mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
1694
1695 #ifdef CONFIG_MLX4_EN_VXLAN
1696         if (priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1697                 vxlan_get_rx_port(dev);
1698 #endif
1699         priv->port_up = true;
1700         netif_tx_start_all_queues(dev);
1701         netif_device_attach(dev);
1702
1703         return 0;
1704
1705 tx_err:
1706         while (tx_index--) {
1707                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1708                 mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1709         }
1710         mlx4_en_destroy_drop_qp(priv);
1711 rss_err:
1712         mlx4_en_release_rss_steer(priv);
1713 mac_err:
1714         mlx4_en_put_qp(priv);
1715 cq_err:
1716         while (rx_index--) {
1717                 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1718                 mlx4_en_free_affinity_hint(priv, i);
1719         }
1720         for (i = 0; i < priv->rx_ring_num; i++)
1721                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1722
1723         return err; /* need to close devices */
1724 }
1725
1726
1727 void mlx4_en_stop_port(struct net_device *dev, int detach)
1728 {
1729         struct mlx4_en_priv *priv = netdev_priv(dev);
1730         struct mlx4_en_dev *mdev = priv->mdev;
1731         struct mlx4_en_mc_list *mclist, *tmp;
1732         struct ethtool_flow_id *flow, *tmp_flow;
1733         int i;
1734         u8 mc_list[16] = {0};
1735
1736         if (!priv->port_up) {
1737                 en_dbg(DRV, priv, "stop port called while port already down\n");
1738                 return;
1739         }
1740
1741         /* close port*/
1742         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1743
1744         /* Synchronize with tx routine */
1745         netif_tx_lock_bh(dev);
1746         if (detach)
1747                 netif_device_detach(dev);
1748         netif_tx_stop_all_queues(dev);
1749         netif_tx_unlock_bh(dev);
1750
1751         netif_tx_disable(dev);
1752
1753         /* Set port as not active */
1754         priv->port_up = false;
1755
1756         /* Promsicuous mode */
1757         if (mdev->dev->caps.steering_mode ==
1758             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1759                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1760                                  MLX4_EN_FLAG_MC_PROMISC);
1761                 mlx4_flow_steer_promisc_remove(mdev->dev,
1762                                                priv->port,
1763                                                MLX4_FS_ALL_DEFAULT);
1764                 mlx4_flow_steer_promisc_remove(mdev->dev,
1765                                                priv->port,
1766                                                MLX4_FS_MC_DEFAULT);
1767         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1768                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1769
1770                 /* Disable promiscouos mode */
1771                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1772                                             priv->port);
1773
1774                 /* Disable Multicast promisc */
1775                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1776                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1777                                                       priv->port);
1778                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1779                 }
1780         }
1781
1782         /* Detach All multicasts */
1783         memset(&mc_list[10], 0xff, ETH_ALEN);
1784         mc_list[5] = priv->port; /* needed for B0 steering support */
1785         mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1786                               MLX4_PROT_ETH, priv->broadcast_id);
1787         list_for_each_entry(mclist, &priv->curr_list, list) {
1788                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1789                 mc_list[5] = priv->port;
1790                 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1791                                       mc_list, MLX4_PROT_ETH, mclist->reg_id);
1792                 if (mclist->tunnel_reg_id)
1793                         mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1794         }
1795         mlx4_en_clear_list(dev);
1796         list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1797                 list_del(&mclist->list);
1798                 kfree(mclist);
1799         }
1800
1801         /* Flush multicast filter */
1802         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1803
1804         /* Remove flow steering rules for the port*/
1805         if (mdev->dev->caps.steering_mode ==
1806             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1807                 ASSERT_RTNL();
1808                 list_for_each_entry_safe(flow, tmp_flow,
1809                                          &priv->ethtool_list, list) {
1810                         mlx4_flow_detach(mdev->dev, flow->id);
1811                         list_del(&flow->list);
1812                 }
1813         }
1814
1815         mlx4_en_destroy_drop_qp(priv);
1816
1817         /* Free TX Rings */
1818         for (i = 0; i < priv->tx_ring_num; i++) {
1819                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1820                 mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1821         }
1822         msleep(10);
1823
1824         for (i = 0; i < priv->tx_ring_num; i++)
1825                 mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1826
1827         /* Free RSS qps */
1828         mlx4_en_release_rss_steer(priv);
1829
1830         /* Unregister Mac address for the port */
1831         mlx4_en_put_qp(priv);
1832         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1833                 mdev->mac_removed[priv->port] = 1;
1834
1835         /* Free RX Rings */
1836         for (i = 0; i < priv->rx_ring_num; i++) {
1837                 struct mlx4_en_cq *cq = priv->rx_cq[i];
1838
1839                 local_bh_disable();
1840                 while (!mlx4_en_cq_lock_napi(cq)) {
1841                         pr_info("CQ %d locked\n", i);
1842                         mdelay(1);
1843                 }
1844                 local_bh_enable();
1845
1846                 napi_synchronize(&cq->napi);
1847                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1848                 mlx4_en_deactivate_cq(priv, cq);
1849
1850                 mlx4_en_free_affinity_hint(priv, i);
1851         }
1852 }
1853
1854 static void mlx4_en_restart(struct work_struct *work)
1855 {
1856         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1857                                                  watchdog_task);
1858         struct mlx4_en_dev *mdev = priv->mdev;
1859         struct net_device *dev = priv->dev;
1860
1861         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1862
1863         mutex_lock(&mdev->state_lock);
1864         if (priv->port_up) {
1865                 mlx4_en_stop_port(dev, 1);
1866                 if (mlx4_en_start_port(dev))
1867                         en_err(priv, "Failed restarting port %d\n", priv->port);
1868         }
1869         mutex_unlock(&mdev->state_lock);
1870 }
1871
1872 static void mlx4_en_clear_stats(struct net_device *dev)
1873 {
1874         struct mlx4_en_priv *priv = netdev_priv(dev);
1875         struct mlx4_en_dev *mdev = priv->mdev;
1876         int i;
1877
1878         if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1879                 en_dbg(HW, priv, "Failed dumping statistics\n");
1880
1881         memset(&priv->stats, 0, sizeof(priv->stats));
1882         memset(&priv->pstats, 0, sizeof(priv->pstats));
1883         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1884         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1885
1886         for (i = 0; i < priv->tx_ring_num; i++) {
1887                 priv->tx_ring[i]->bytes = 0;
1888                 priv->tx_ring[i]->packets = 0;
1889                 priv->tx_ring[i]->tx_csum = 0;
1890         }
1891         for (i = 0; i < priv->rx_ring_num; i++) {
1892                 priv->rx_ring[i]->bytes = 0;
1893                 priv->rx_ring[i]->packets = 0;
1894                 priv->rx_ring[i]->csum_ok = 0;
1895                 priv->rx_ring[i]->csum_none = 0;
1896                 priv->rx_ring[i]->csum_complete = 0;
1897         }
1898 }
1899
1900 static int mlx4_en_open(struct net_device *dev)
1901 {
1902         struct mlx4_en_priv *priv = netdev_priv(dev);
1903         struct mlx4_en_dev *mdev = priv->mdev;
1904         int err = 0;
1905
1906         mutex_lock(&mdev->state_lock);
1907
1908         if (!mdev->device_up) {
1909                 en_err(priv, "Cannot open - device down/disabled\n");
1910                 err = -EBUSY;
1911                 goto out;
1912         }
1913
1914         /* Reset HW statistics and SW counters */
1915         mlx4_en_clear_stats(dev);
1916
1917         err = mlx4_en_start_port(dev);
1918         if (err)
1919                 en_err(priv, "Failed starting port:%d\n", priv->port);
1920
1921 out:
1922         mutex_unlock(&mdev->state_lock);
1923         return err;
1924 }
1925
1926
1927 static int mlx4_en_close(struct net_device *dev)
1928 {
1929         struct mlx4_en_priv *priv = netdev_priv(dev);
1930         struct mlx4_en_dev *mdev = priv->mdev;
1931
1932         en_dbg(IFDOWN, priv, "Close port called\n");
1933
1934         mutex_lock(&mdev->state_lock);
1935
1936         mlx4_en_stop_port(dev, 0);
1937         netif_carrier_off(dev);
1938
1939         mutex_unlock(&mdev->state_lock);
1940         return 0;
1941 }
1942
1943 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1944 {
1945         int i;
1946
1947 #ifdef CONFIG_RFS_ACCEL
1948         free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1949         priv->dev->rx_cpu_rmap = NULL;
1950 #endif
1951
1952         for (i = 0; i < priv->tx_ring_num; i++) {
1953                 if (priv->tx_ring && priv->tx_ring[i])
1954                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1955                 if (priv->tx_cq && priv->tx_cq[i])
1956                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1957         }
1958
1959         for (i = 0; i < priv->rx_ring_num; i++) {
1960                 if (priv->rx_ring[i])
1961                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1962                                 priv->prof->rx_ring_size, priv->stride);
1963                 if (priv->rx_cq[i])
1964                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1965         }
1966
1967         if (priv->base_tx_qpn) {
1968                 mlx4_qp_release_range(priv->mdev->dev, priv->base_tx_qpn, priv->tx_ring_num);
1969                 priv->base_tx_qpn = 0;
1970         }
1971 }
1972
1973 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1974 {
1975         struct mlx4_en_port_profile *prof = priv->prof;
1976         int i;
1977         int err;
1978         int node;
1979
1980         err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &priv->base_tx_qpn);
1981         if (err) {
1982                 en_err(priv, "failed reserving range for TX rings\n");
1983                 return err;
1984         }
1985
1986         /* Create tx Rings */
1987         for (i = 0; i < priv->tx_ring_num; i++) {
1988                 node = cpu_to_node(i % num_online_cpus());
1989                 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1990                                       prof->tx_ring_size, i, TX, node))
1991                         goto err;
1992
1993                 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
1994                                            priv->base_tx_qpn + i,
1995                                            prof->tx_ring_size, TXBB_SIZE,
1996                                            node, i))
1997                         goto err;
1998         }
1999
2000         /* Create rx Rings */
2001         for (i = 0; i < priv->rx_ring_num; i++) {
2002                 node = cpu_to_node(i % num_online_cpus());
2003                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2004                                       prof->rx_ring_size, i, RX, node))
2005                         goto err;
2006
2007                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2008                                            prof->rx_ring_size, priv->stride,
2009                                            node))
2010                         goto err;
2011         }
2012
2013 #ifdef CONFIG_RFS_ACCEL
2014         if (priv->mdev->dev->caps.comp_pool) {
2015                 priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->mdev->dev->caps.comp_pool);
2016                 if (!priv->dev->rx_cpu_rmap)
2017                         goto err;
2018         }
2019 #endif
2020
2021         return 0;
2022
2023 err:
2024         en_err(priv, "Failed to allocate NIC resources\n");
2025         for (i = 0; i < priv->rx_ring_num; i++) {
2026                 if (priv->rx_ring[i])
2027                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2028                                                 prof->rx_ring_size,
2029                                                 priv->stride);
2030                 if (priv->rx_cq[i])
2031                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2032         }
2033         for (i = 0; i < priv->tx_ring_num; i++) {
2034                 if (priv->tx_ring[i])
2035                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
2036                 if (priv->tx_cq[i])
2037                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
2038         }
2039         return -ENOMEM;
2040 }
2041
2042
2043 void mlx4_en_destroy_netdev(struct net_device *dev)
2044 {
2045         struct mlx4_en_priv *priv = netdev_priv(dev);
2046         struct mlx4_en_dev *mdev = priv->mdev;
2047
2048         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2049
2050         /* Unregister device - this will close the port if it was up */
2051         if (priv->registered)
2052                 unregister_netdev(dev);
2053
2054         if (priv->allocated)
2055                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2056
2057         cancel_delayed_work(&priv->stats_task);
2058         cancel_delayed_work(&priv->service_task);
2059         /* flush any pending task for this netdev */
2060         flush_workqueue(mdev->workqueue);
2061
2062         /* Detach the netdev so tasks would not attempt to access it */
2063         mutex_lock(&mdev->state_lock);
2064         mdev->pndev[priv->port] = NULL;
2065         mutex_unlock(&mdev->state_lock);
2066
2067         mlx4_en_free_resources(priv);
2068
2069         kfree(priv->tx_ring);
2070         kfree(priv->tx_cq);
2071
2072         free_netdev(dev);
2073 }
2074
2075 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2076 {
2077         struct mlx4_en_priv *priv = netdev_priv(dev);
2078         struct mlx4_en_dev *mdev = priv->mdev;
2079         int err = 0;
2080
2081         en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2082                  dev->mtu, new_mtu);
2083
2084         if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
2085                 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
2086                 return -EPERM;
2087         }
2088         dev->mtu = new_mtu;
2089
2090         if (netif_running(dev)) {
2091                 mutex_lock(&mdev->state_lock);
2092                 if (!mdev->device_up) {
2093                         /* NIC is probably restarting - let watchdog task reset
2094                          * the port */
2095                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2096                 } else {
2097                         mlx4_en_stop_port(dev, 1);
2098                         err = mlx4_en_start_port(dev);
2099                         if (err) {
2100                                 en_err(priv, "Failed restarting port:%d\n",
2101                                          priv->port);
2102                                 queue_work(mdev->workqueue, &priv->watchdog_task);
2103                         }
2104                 }
2105                 mutex_unlock(&mdev->state_lock);
2106         }
2107         return 0;
2108 }
2109
2110 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2111 {
2112         struct mlx4_en_priv *priv = netdev_priv(dev);
2113         struct mlx4_en_dev *mdev = priv->mdev;
2114         struct hwtstamp_config config;
2115
2116         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2117                 return -EFAULT;
2118
2119         /* reserved for future extensions */
2120         if (config.flags)
2121                 return -EINVAL;
2122
2123         /* device doesn't support time stamping */
2124         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2125                 return -EINVAL;
2126
2127         /* TX HW timestamp */
2128         switch (config.tx_type) {
2129         case HWTSTAMP_TX_OFF:
2130         case HWTSTAMP_TX_ON:
2131                 break;
2132         default:
2133                 return -ERANGE;
2134         }
2135
2136         /* RX HW timestamp */
2137         switch (config.rx_filter) {
2138         case HWTSTAMP_FILTER_NONE:
2139                 break;
2140         case HWTSTAMP_FILTER_ALL:
2141         case HWTSTAMP_FILTER_SOME:
2142         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2143         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2144         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2145         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2146         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2147         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2148         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2149         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2150         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2151         case HWTSTAMP_FILTER_PTP_V2_EVENT:
2152         case HWTSTAMP_FILTER_PTP_V2_SYNC:
2153         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2154                 config.rx_filter = HWTSTAMP_FILTER_ALL;
2155                 break;
2156         default:
2157                 return -ERANGE;
2158         }
2159
2160         if (mlx4_en_reset_config(dev, config, dev->features)) {
2161                 config.tx_type = HWTSTAMP_TX_OFF;
2162                 config.rx_filter = HWTSTAMP_FILTER_NONE;
2163         }
2164
2165         return copy_to_user(ifr->ifr_data, &config,
2166                             sizeof(config)) ? -EFAULT : 0;
2167 }
2168
2169 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2170 {
2171         struct mlx4_en_priv *priv = netdev_priv(dev);
2172
2173         return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2174                             sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2175 }
2176
2177 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2178 {
2179         switch (cmd) {
2180         case SIOCSHWTSTAMP:
2181                 return mlx4_en_hwtstamp_set(dev, ifr);
2182         case SIOCGHWTSTAMP:
2183                 return mlx4_en_hwtstamp_get(dev, ifr);
2184         default:
2185                 return -EOPNOTSUPP;
2186         }
2187 }
2188
2189 static int mlx4_en_set_features(struct net_device *netdev,
2190                 netdev_features_t features)
2191 {
2192         struct mlx4_en_priv *priv = netdev_priv(netdev);
2193         int ret = 0;
2194
2195         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2196                 en_info(priv, "Turn %s RX vlan strip offload\n",
2197                         (features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2198                 ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2199                                            features);
2200                 if (ret)
2201                         return ret;
2202         }
2203
2204         if (features & NETIF_F_LOOPBACK)
2205                 priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
2206         else
2207                 priv->ctrl_flags &=
2208                         cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
2209
2210         mlx4_en_update_loopback_state(netdev, features);
2211
2212         return 0;
2213
2214 }
2215
2216 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2217 {
2218         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2219         struct mlx4_en_dev *mdev = en_priv->mdev;
2220         u64 mac_u64 = mlx4_mac_to_u64(mac);
2221
2222         if (!is_valid_ether_addr(mac))
2223                 return -EINVAL;
2224
2225         return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2226 }
2227
2228 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos)
2229 {
2230         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2231         struct mlx4_en_dev *mdev = en_priv->mdev;
2232
2233         return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos);
2234 }
2235
2236 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2237 {
2238         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2239         struct mlx4_en_dev *mdev = en_priv->mdev;
2240
2241         return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2242 }
2243
2244 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2245 {
2246         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2247         struct mlx4_en_dev *mdev = en_priv->mdev;
2248
2249         return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2250 }
2251
2252 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2253 {
2254         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2255         struct mlx4_en_dev *mdev = en_priv->mdev;
2256
2257         return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2258 }
2259
2260 #define PORT_ID_BYTE_LEN 8
2261 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2262                                     struct netdev_phys_port_id *ppid)
2263 {
2264         struct mlx4_en_priv *priv = netdev_priv(dev);
2265         struct mlx4_dev *mdev = priv->mdev->dev;
2266         int i;
2267         u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2268
2269         if (!phys_port_id)
2270                 return -EOPNOTSUPP;
2271
2272         ppid->id_len = sizeof(phys_port_id);
2273         for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2274                 ppid->id[i] =  phys_port_id & 0xff;
2275                 phys_port_id >>= 8;
2276         }
2277         return 0;
2278 }
2279
2280 #ifdef CONFIG_MLX4_EN_VXLAN
2281 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2282 {
2283         int ret;
2284         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2285                                                  vxlan_add_task);
2286
2287         ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2288         if (ret)
2289                 goto out;
2290
2291         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2292                                   VXLAN_STEER_BY_OUTER_MAC, 1);
2293 out:
2294         if (ret) {
2295                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2296                 return;
2297         }
2298
2299         /* set offloads */
2300         priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
2301                                       NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL;
2302         priv->dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
2303         priv->dev->features    |= NETIF_F_GSO_UDP_TUNNEL;
2304 }
2305
2306 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2307 {
2308         int ret;
2309         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2310                                                  vxlan_del_task);
2311         /* unset offloads */
2312         priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
2313                                       NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL);
2314         priv->dev->hw_features &= ~NETIF_F_GSO_UDP_TUNNEL;
2315         priv->dev->features    &= ~NETIF_F_GSO_UDP_TUNNEL;
2316
2317         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2318                                   VXLAN_STEER_BY_OUTER_MAC, 0);
2319         if (ret)
2320                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2321
2322         priv->vxlan_port = 0;
2323 }
2324
2325 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2326                                    sa_family_t sa_family, __be16 port)
2327 {
2328         struct mlx4_en_priv *priv = netdev_priv(dev);
2329         __be16 current_port;
2330
2331         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2332                 return;
2333
2334         if (sa_family == AF_INET6)
2335                 return;
2336
2337         current_port = priv->vxlan_port;
2338         if (current_port && current_port != port) {
2339                 en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2340                         ntohs(current_port), ntohs(port));
2341                 return;
2342         }
2343
2344         priv->vxlan_port = port;
2345         queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2346 }
2347
2348 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2349                                    sa_family_t sa_family, __be16 port)
2350 {
2351         struct mlx4_en_priv *priv = netdev_priv(dev);
2352         __be16 current_port;
2353
2354         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2355                 return;
2356
2357         if (sa_family == AF_INET6)
2358                 return;
2359
2360         current_port = priv->vxlan_port;
2361         if (current_port != port) {
2362                 en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2363                 return;
2364         }
2365
2366         queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2367 }
2368 #endif
2369
2370 static const struct net_device_ops mlx4_netdev_ops = {
2371         .ndo_open               = mlx4_en_open,
2372         .ndo_stop               = mlx4_en_close,
2373         .ndo_start_xmit         = mlx4_en_xmit,
2374         .ndo_select_queue       = mlx4_en_select_queue,
2375         .ndo_get_stats          = mlx4_en_get_stats,
2376         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2377         .ndo_set_mac_address    = mlx4_en_set_mac,
2378         .ndo_validate_addr      = eth_validate_addr,
2379         .ndo_change_mtu         = mlx4_en_change_mtu,
2380         .ndo_do_ioctl           = mlx4_en_ioctl,
2381         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2382         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2383         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2384 #ifdef CONFIG_NET_POLL_CONTROLLER
2385         .ndo_poll_controller    = mlx4_en_netpoll,
2386 #endif
2387         .ndo_set_features       = mlx4_en_set_features,
2388         .ndo_setup_tc           = mlx4_en_setup_tc,
2389 #ifdef CONFIG_RFS_ACCEL
2390         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2391 #endif
2392 #ifdef CONFIG_NET_RX_BUSY_POLL
2393         .ndo_busy_poll          = mlx4_en_low_latency_recv,
2394 #endif
2395         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2396 #ifdef CONFIG_MLX4_EN_VXLAN
2397         .ndo_add_vxlan_port     = mlx4_en_add_vxlan_port,
2398         .ndo_del_vxlan_port     = mlx4_en_del_vxlan_port,
2399 #endif
2400 };
2401
2402 static const struct net_device_ops mlx4_netdev_ops_master = {
2403         .ndo_open               = mlx4_en_open,
2404         .ndo_stop               = mlx4_en_close,
2405         .ndo_start_xmit         = mlx4_en_xmit,
2406         .ndo_select_queue       = mlx4_en_select_queue,
2407         .ndo_get_stats          = mlx4_en_get_stats,
2408         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2409         .ndo_set_mac_address    = mlx4_en_set_mac,
2410         .ndo_validate_addr      = eth_validate_addr,
2411         .ndo_change_mtu         = mlx4_en_change_mtu,
2412         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2413         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2414         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2415         .ndo_set_vf_mac         = mlx4_en_set_vf_mac,
2416         .ndo_set_vf_vlan        = mlx4_en_set_vf_vlan,
2417         .ndo_set_vf_spoofchk    = mlx4_en_set_vf_spoofchk,
2418         .ndo_set_vf_link_state  = mlx4_en_set_vf_link_state,
2419         .ndo_get_vf_config      = mlx4_en_get_vf_config,
2420 #ifdef CONFIG_NET_POLL_CONTROLLER
2421         .ndo_poll_controller    = mlx4_en_netpoll,
2422 #endif
2423         .ndo_set_features       = mlx4_en_set_features,
2424         .ndo_setup_tc           = mlx4_en_setup_tc,
2425 #ifdef CONFIG_RFS_ACCEL
2426         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2427 #endif
2428         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2429 };
2430
2431 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2432                         struct mlx4_en_port_profile *prof)
2433 {
2434         struct net_device *dev;
2435         struct mlx4_en_priv *priv;
2436         int i;
2437         int err;
2438         u64 mac_u64;
2439
2440         dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2441                                  MAX_TX_RINGS, MAX_RX_RINGS);
2442         if (dev == NULL)
2443                 return -ENOMEM;
2444
2445         netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2446         netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2447
2448         SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
2449         dev->dev_port = port - 1;
2450
2451         /*
2452          * Initialize driver private data
2453          */
2454
2455         priv = netdev_priv(dev);
2456         memset(priv, 0, sizeof(struct mlx4_en_priv));
2457         spin_lock_init(&priv->stats_lock);
2458         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2459         INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2460         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2461         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2462         INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2463 #ifdef CONFIG_MLX4_EN_VXLAN
2464         INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
2465         INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
2466 #endif
2467 #ifdef CONFIG_RFS_ACCEL
2468         INIT_LIST_HEAD(&priv->filters);
2469         spin_lock_init(&priv->filters_lock);
2470 #endif
2471
2472         priv->dev = dev;
2473         priv->mdev = mdev;
2474         priv->ddev = &mdev->pdev->dev;
2475         priv->prof = prof;
2476         priv->port = port;
2477         priv->port_up = false;
2478         priv->flags = prof->flags;
2479         priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
2480         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2481                         MLX4_WQE_CTRL_SOLICITED);
2482         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2483         priv->tx_ring_num = prof->tx_ring_num;
2484         priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
2485
2486         priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2487                                 GFP_KERNEL);
2488         if (!priv->tx_ring) {
2489                 err = -ENOMEM;
2490                 goto out;
2491         }
2492         priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2493                               GFP_KERNEL);
2494         if (!priv->tx_cq) {
2495                 err = -ENOMEM;
2496                 goto out;
2497         }
2498         priv->rx_ring_num = prof->rx_ring_num;
2499         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2500         priv->cqe_size = mdev->dev->caps.cqe_size;
2501         priv->mac_index = -1;
2502         priv->msg_enable = MLX4_EN_MSG_LEVEL;
2503 #ifdef CONFIG_MLX4_EN_DCB
2504         if (!mlx4_is_slave(priv->mdev->dev)) {
2505                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
2506                         dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2507                 } else {
2508                         en_info(priv, "enabling only PFC DCB ops\n");
2509                         dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2510                 }
2511         }
2512 #endif
2513
2514         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2515                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2516
2517         /* Query for default mac and max mtu */
2518         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2519
2520         if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
2521             MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
2522                 priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
2523
2524         /* Set default MAC */
2525         dev->addr_len = ETH_ALEN;
2526         mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2527         if (!is_valid_ether_addr(dev->dev_addr)) {
2528                 if (mlx4_is_slave(priv->mdev->dev)) {
2529                         eth_hw_addr_random(dev);
2530                         en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
2531                         mac_u64 = mlx4_mac_to_u64(dev->dev_addr);
2532                         mdev->dev->caps.def_mac[priv->port] = mac_u64;
2533                 } else {
2534                         en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2535                                priv->port, dev->dev_addr);
2536                         err = -EINVAL;
2537                         goto out;
2538                 }
2539         }
2540
2541         memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
2542
2543         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2544                                           DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2545         err = mlx4_en_alloc_resources(priv);
2546         if (err)
2547                 goto out;
2548
2549         /* Initialize time stamping config */
2550         priv->hwtstamp_config.flags = 0;
2551         priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
2552         priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2553
2554         /* Allocate page for receive rings */
2555         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2556                                 MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2557         if (err) {
2558                 en_err(priv, "Failed to allocate page for rx qps\n");
2559                 goto out;
2560         }
2561         priv->allocated = 1;
2562
2563         /*
2564          * Initialize netdev entry points
2565          */
2566         if (mlx4_is_master(priv->mdev->dev))
2567                 dev->netdev_ops = &mlx4_netdev_ops_master;
2568         else
2569                 dev->netdev_ops = &mlx4_netdev_ops;
2570         dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2571         netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2572         netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2573
2574         dev->ethtool_ops = &mlx4_en_ethtool_ops;
2575
2576         /*
2577          * Set driver features
2578          */
2579         dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2580         if (mdev->LSO_support)
2581                 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2582
2583         dev->vlan_features = dev->hw_features;
2584
2585         dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2586         dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2587                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2588                         NETIF_F_HW_VLAN_CTAG_FILTER;
2589         dev->hw_features |= NETIF_F_LOOPBACK |
2590                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2591
2592         if (mdev->dev->caps.steering_mode ==
2593             MLX4_STEERING_MODE_DEVICE_MANAGED)
2594                 dev->hw_features |= NETIF_F_NTUPLE;
2595
2596         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
2597                 dev->priv_flags |= IFF_UNICAST_FLT;
2598
2599         mdev->pndev[port] = dev;
2600
2601         netif_carrier_off(dev);
2602         mlx4_en_set_default_moderation(priv);
2603
2604         err = register_netdev(dev);
2605         if (err) {
2606                 en_err(priv, "Netdev registration failed for port %d\n", port);
2607                 goto out;
2608         }
2609         priv->registered = 1;
2610
2611         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2612         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2613
2614         mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
2615
2616         /* Configure port */
2617         mlx4_en_calc_rx_buf(dev);
2618         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2619                                     priv->rx_skb_size + ETH_FCS_LEN,
2620                                     prof->tx_pause, prof->tx_ppp,
2621                                     prof->rx_pause, prof->rx_ppp);
2622         if (err) {
2623                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
2624                        priv->port, err);
2625                 goto out;
2626         }
2627
2628         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2629                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
2630                 if (err) {
2631                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
2632                                err);
2633                         goto out;
2634                 }
2635         }
2636
2637         /* Init port */
2638         en_warn(priv, "Initializing port\n");
2639         err = mlx4_INIT_PORT(mdev->dev, priv->port);
2640         if (err) {
2641                 en_err(priv, "Failed Initializing port\n");
2642                 goto out;
2643         }
2644         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2645
2646         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2647                 queue_delayed_work(mdev->workqueue, &priv->service_task,
2648                                    SERVICE_TASK_DELAY);
2649
2650         return 0;
2651
2652 out:
2653         mlx4_en_destroy_netdev(dev);
2654         return err;
2655 }
2656
2657 int mlx4_en_reset_config(struct net_device *dev,
2658                          struct hwtstamp_config ts_config,
2659                          netdev_features_t features)
2660 {
2661         struct mlx4_en_priv *priv = netdev_priv(dev);
2662         struct mlx4_en_dev *mdev = priv->mdev;
2663         int port_up = 0;
2664         int err = 0;
2665
2666         if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
2667             priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
2668             !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX))
2669                 return 0; /* Nothing to change */
2670
2671         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
2672             (features & NETIF_F_HW_VLAN_CTAG_RX) &&
2673             (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
2674                 en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
2675                 return -EINVAL;
2676         }
2677
2678         mutex_lock(&mdev->state_lock);
2679         if (priv->port_up) {
2680                 port_up = 1;
2681                 mlx4_en_stop_port(dev, 1);
2682         }
2683
2684         mlx4_en_free_resources(priv);
2685
2686         en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
2687                 ts_config.rx_filter, !!(features & NETIF_F_HW_VLAN_CTAG_RX));
2688
2689         priv->hwtstamp_config.tx_type = ts_config.tx_type;
2690         priv->hwtstamp_config.rx_filter = ts_config.rx_filter;
2691
2692         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2693                 if (features & NETIF_F_HW_VLAN_CTAG_RX)
2694                         dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
2695                 else
2696                         dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
2697         } else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
2698                 /* RX time-stamping is OFF, update the RX vlan offload
2699                  * to the latest wanted state
2700                  */
2701                 if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
2702                         dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
2703                 else
2704                         dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
2705         }
2706
2707         /* RX vlan offload and RX time-stamping can't co-exist !
2708          * Regardless of the caller's choice,
2709          * Turn Off RX vlan offload in case of time-stamping is ON
2710          */
2711         if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
2712                 if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
2713                         en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
2714                 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
2715         }
2716
2717         err = mlx4_en_alloc_resources(priv);
2718         if (err) {
2719                 en_err(priv, "Failed reallocating port resources\n");
2720                 goto out;
2721         }
2722         if (port_up) {
2723                 err = mlx4_en_start_port(dev);
2724                 if (err)
2725                         en_err(priv, "Failed starting port\n");
2726         }
2727
2728 out:
2729         mutex_unlock(&mdev->state_lock);
2730         netdev_features_change(dev);
2731         return err;
2732 }