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
  43 #include <linux/mlx4/driver.h>
  44 #include <linux/mlx4/device.h>
  45 #include <linux/mlx4/cmd.h>
  46 #include <linux/mlx4/cq.h>
  47
  48 #include "mlx4_en.h"
  49 #include "en_port.h"
  50
  51 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
  52 {
  53         struct mlx4_en_priv *priv = netdev_priv(dev);
  54         int i;
  55         unsigned int offset = 0;
  56
  57         if (up && up != MLX4_EN_NUM_UP)
  58                 return -EINVAL;
  59
  60         netdev_set_num_tc(dev, up);
  61
  62         /* Partition Tx queues evenly amongst UP's */
  63         for (i = 0; i < up; i++) {
  64                 netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
  65                 offset += priv->num_tx_rings_p_up;
  66         }
  67
  68         return 0;
  69 }
  70
  71 #ifdef CONFIG_NET_RX_BUSY_POLL
  72 /* must be called with local_bh_disable()d */
  73 static int mlx4_en_low_latency_recv(struct napi_struct *napi)
  74 {
  75         struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
  76         struct net_device *dev = cq->dev;
  77         struct mlx4_en_priv *priv = netdev_priv(dev);
  78         struct mlx4_en_rx_ring *rx_ring = priv->rx_ring[cq->ring];
  79         int done;
  80
  81         if (!priv->port_up)
  82                 return LL_FLUSH_FAILED;
  83
  84         if (!mlx4_en_cq_lock_poll(cq))
  85                 return LL_FLUSH_BUSY;
  86
  87         done = mlx4_en_process_rx_cq(dev, cq, 4);
  88         if (likely(done))
  89                 rx_ring->cleaned += done;
  90         else
  91                 rx_ring->misses++;
  92
  93         mlx4_en_cq_unlock_poll(cq);
  94
  95         return done;
  96 }
  97 #endif  /* CONFIG_NET_RX_BUSY_POLL */
  98
  99 #ifdef CONFIG_RFS_ACCEL
 100
 101 struct mlx4_en_filter {
 102         struct list_head next;
 103         struct work_struct work;
 104
 105         u8     ip_proto;
 106         __be32 src_ip;
 107         __be32 dst_ip;
 108         __be16 src_port;
 109         __be16 dst_port;
 110
 111         int rxq_index;
 112         struct mlx4_en_priv *priv;
 113         u32 flow_id;                    /* RFS infrastructure id */
 114         int id;                         /* mlx4_en driver id */
 115         u64 reg_id;                     /* Flow steering API id */
 116         u8 activated;                   /* Used to prevent expiry before filter
 117                                          * is attached
 118                                          */
 119         struct hlist_node filter_chain;
 120 };
 121
 122 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
 123
 124 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
 125 {
 126         switch (ip_proto) {
 127         case IPPROTO_UDP:
 128                 return MLX4_NET_TRANS_RULE_ID_UDP;
 129         case IPPROTO_TCP:
 130                 return MLX4_NET_TRANS_RULE_ID_TCP;
 131         default:
 132                 return -EPROTONOSUPPORT;
 133         }
 134 };
 135
 136 static void mlx4_en_filter_work(struct work_struct *work)
 137 {
 138         struct mlx4_en_filter *filter = container_of(work,
 139                                                      struct mlx4_en_filter,
 140                                                      work);
 141         struct mlx4_en_priv *priv = filter->priv;
 142         struct mlx4_spec_list spec_tcp_udp = {
 143                 .id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
 144                 {
 145                         .tcp_udp = {
 146                                 .dst_port = filter->dst_port,
 147                                 .dst_port_msk = (__force __be16)-1,
 148                                 .src_port = filter->src_port,
 149                                 .src_port_msk = (__force __be16)-1,
 150                         },
 151                 },
 152         };
 153         struct mlx4_spec_list spec_ip = {
 154                 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
 155                 {
 156                         .ipv4 = {
 157                                 .dst_ip = filter->dst_ip,
 158                                 .dst_ip_msk = (__force __be32)-1,
 159                                 .src_ip = filter->src_ip,
 160                                 .src_ip_msk = (__force __be32)-1,
 161                         },
 162                 },
 163         };
 164         struct mlx4_spec_list spec_eth = {
 165                 .id = MLX4_NET_TRANS_RULE_ID_ETH,
 166         };
 167         struct mlx4_net_trans_rule rule = {
 168                 .list = LIST_HEAD_INIT(rule.list),
 169                 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
 170                 .exclusive = 1,
 171                 .allow_loopback = 1,
 172                 .promisc_mode = MLX4_FS_REGULAR,
 173                 .port = priv->port,
 174                 .priority = MLX4_DOMAIN_RFS,
 175         };
 176         int rc;
 177         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 178
 179         if (spec_tcp_udp.id < 0) {
 180                 en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
 181                         filter->ip_proto);
 182                 goto ignore;
 183         }
 184         list_add_tail(&spec_eth.list, &rule.list);
 185         list_add_tail(&spec_ip.list, &rule.list);
 186         list_add_tail(&spec_tcp_udp.list, &rule.list);
 187
 188         rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
 189         memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
 190         memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 191
 192         filter->activated = 0;
 193
 194         if (filter->reg_id) {
 195                 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 196                 if (rc && rc != -ENOENT)
 197                         en_err(priv, "Error detaching flow. rc = %d\n", rc);
 198         }
 199
 200         rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
 201         if (rc)
 202                 en_err(priv, "Error attaching flow. err = %d\n", rc);
 203
 204 ignore:
 205         mlx4_en_filter_rfs_expire(priv);
 206
 207         filter->activated = 1;
 208 }
 209
 210 static inline struct hlist_head *
 211 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 212                    __be16 src_port, __be16 dst_port)
 213 {
 214         unsigned long l;
 215         int bucket_idx;
 216
 217         l = (__force unsigned long)src_port |
 218             ((__force unsigned long)dst_port << 2);
 219         l ^= (__force unsigned long)(src_ip ^ dst_ip);
 220
 221         bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
 222
 223         return &priv->filter_hash[bucket_idx];
 224 }
 225
 226 static struct mlx4_en_filter *
 227 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
 228                      __be32 dst_ip, u8 ip_proto, __be16 src_port,
 229                      __be16 dst_port, u32 flow_id)
 230 {
 231         struct mlx4_en_filter *filter = NULL;
 232
 233         filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
 234         if (!filter)
 235                 return NULL;
 236
 237         filter->priv = priv;
 238         filter->rxq_index = rxq_index;
 239         INIT_WORK(&filter->work, mlx4_en_filter_work);
 240
 241         filter->src_ip = src_ip;
 242         filter->dst_ip = dst_ip;
 243         filter->ip_proto = ip_proto;
 244         filter->src_port = src_port;
 245         filter->dst_port = dst_port;
 246
 247         filter->flow_id = flow_id;
 248
 249         filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
 250
 251         list_add_tail(&filter->next, &priv->filters);
 252         hlist_add_head(&filter->filter_chain,
 253                        filter_hash_bucket(priv, src_ip, dst_ip, src_port,
 254                                           dst_port));
 255
 256         return filter;
 257 }
 258
 259 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
 260 {
 261         struct mlx4_en_priv *priv = filter->priv;
 262         int rc;
 263
 264         list_del(&filter->next);
 265
 266         rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 267         if (rc && rc != -ENOENT)
 268                 en_err(priv, "Error detaching flow. rc = %d\n", rc);
 269
 270         kfree(filter);
 271 }
 272
 273 static inline struct mlx4_en_filter *
 274 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 275                     u8 ip_proto, __be16 src_port, __be16 dst_port)
 276 {
 277         struct mlx4_en_filter *filter;
 278         struct mlx4_en_filter *ret = NULL;
 279
 280         hlist_for_each_entry(filter,
 281                              filter_hash_bucket(priv, src_ip, dst_ip,
 282                                                 src_port, dst_port),
 283                              filter_chain) {
 284                 if (filter->src_ip == src_ip &&
 285                     filter->dst_ip == dst_ip &&
 286                     filter->ip_proto == ip_proto &&
 287                     filter->src_port == src_port &&
 288                     filter->dst_port == dst_port) {
 289                         ret = filter;
 290                         break;
 291                 }
 292         }
 293
 294         return ret;
 295 }
 296
 297 static int
 298 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
 299                    u16 rxq_index, u32 flow_id)
 300 {
 301         struct mlx4_en_priv *priv = netdev_priv(net_dev);
 302         struct mlx4_en_filter *filter;
 303         const struct iphdr *ip;
 304         const __be16 *ports;
 305         u8 ip_proto;
 306         __be32 src_ip;
 307         __be32 dst_ip;
 308         __be16 src_port;
 309         __be16 dst_port;
 310         int nhoff = skb_network_offset(skb);
 311         int ret = 0;
 312
 313         if (skb->protocol != htons(ETH_P_IP))
 314                 return -EPROTONOSUPPORT;
 315
 316         ip = (const struct iphdr *)(skb->data + nhoff);
 317         if (ip_is_fragment(ip))
 318                 return -EPROTONOSUPPORT;
 319
 320         if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
 321                 return -EPROTONOSUPPORT;
 322         ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
 323
 324         ip_proto = ip->protocol;
 325         src_ip = ip->saddr;
 326         dst_ip = ip->daddr;
 327         src_port = ports[0];
 328         dst_port = ports[1];
 329
 330         spin_lock_bh(&priv->filters_lock);
 331         filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
 332                                      src_port, dst_port);
 333         if (filter) {
 334                 if (filter->rxq_index == rxq_index)
 335                         goto out;
 336
 337                 filter->rxq_index = rxq_index;
 338         } else {
 339                 filter = mlx4_en_filter_alloc(priv, rxq_index,
 340                                               src_ip, dst_ip, ip_proto,
 341                                               src_port, dst_port, flow_id);
 342                 if (!filter) {
 343                         ret = -ENOMEM;
 344                         goto err;
 345                 }
 346         }
 347
 348         queue_work(priv->mdev->workqueue, &filter->work);
 349
 350 out:
 351         ret = filter->id;
 352 err:
 353         spin_unlock_bh(&priv->filters_lock);
 354
 355         return ret;
 356 }
 357
 358 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
 359 {
 360         struct mlx4_en_filter *filter, *tmp;
 361         LIST_HEAD(del_list);
 362
 363         spin_lock_bh(&priv->filters_lock);
 364         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 365                 list_move(&filter->next, &del_list);
 366                 hlist_del(&filter->filter_chain);
 367         }
 368         spin_unlock_bh(&priv->filters_lock);
 369
 370         list_for_each_entry_safe(filter, tmp, &del_list, next) {
 371                 cancel_work_sync(&filter->work);
 372                 mlx4_en_filter_free(filter);
 373         }
 374 }
 375
 376 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
 377 {
 378         struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
 379         LIST_HEAD(del_list);
 380         int i = 0;
 381
 382         spin_lock_bh(&priv->filters_lock);
 383         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 384                 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
 385                         break;
 386
 387                 if (filter->activated &&
 388                     !work_pending(&filter->work) &&
 389                     rps_may_expire_flow(priv->dev,
 390                                         filter->rxq_index, filter->flow_id,
 391                                         filter->id)) {
 392                         list_move(&filter->next, &del_list);
 393                         hlist_del(&filter->filter_chain);
 394                 } else
 395                         last_filter = filter;
 396
 397                 i++;
 398         }
 399
 400         if (last_filter && (&last_filter->next != priv->filters.next))
 401                 list_move(&priv->filters, &last_filter->next);
 402
 403         spin_unlock_bh(&priv->filters_lock);
 404
 405         list_for_each_entry_safe(filter, tmp, &del_list, next)
 406                 mlx4_en_filter_free(filter);
 407 }
 408 #endif
 409
 410 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
 411                                    __be16 proto, u16 vid)
 412 {
 413         struct mlx4_en_priv *priv = netdev_priv(dev);
 414         struct mlx4_en_dev *mdev = priv->mdev;
 415         int err;
 416         int idx;
 417
 418         en_dbg(HW, priv, "adding VLAN:%d\n", vid);
 419
 420         set_bit(vid, priv->active_vlans);
 421
 422         /* Add VID to port VLAN filter */
 423         mutex_lock(&mdev->state_lock);
 424         if (mdev->device_up && priv->port_up) {
 425                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 426                 if (err)
 427                         en_err(priv, "Failed configuring VLAN filter\n");
 428         }
 429         if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
 430                 en_dbg(HW, priv, "failed adding vlan %d\n", vid);
 431         mutex_unlock(&mdev->state_lock);
 432
 433         return 0;
 434 }
 435
 436 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
 437                                     __be16 proto, u16 vid)
 438 {
 439         struct mlx4_en_priv *priv = netdev_priv(dev);
 440         struct mlx4_en_dev *mdev = priv->mdev;
 441         int err;
 442
 443         en_dbg(HW, priv, "Killing VID:%d\n", vid);
 444
 445         clear_bit(vid, priv->active_vlans);
 446
 447         /* Remove VID from port VLAN filter */
 448         mutex_lock(&mdev->state_lock);
 449         mlx4_unregister_vlan(mdev->dev, priv->port, vid);
 450
 451         if (mdev->device_up && priv->port_up) {
 452                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 453                 if (err)
 454                         en_err(priv, "Failed configuring VLAN filter\n");
 455         }
 456         mutex_unlock(&mdev->state_lock);
 457
 458         return 0;
 459 }
 460
 461 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
 462 {
 463         int i;
 464         for (i = ETH_ALEN - 1; i >= 0; --i) {
 465                 dst_mac[i] = src_mac & 0xff;
 466                 src_mac >>= 8;
 467         }
 468         memset(&dst_mac[ETH_ALEN], 0, 2);
 469 }
 470
 471
 472 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
 473                                     int qpn, u64 *reg_id)
 474 {
 475         int err;
 476         struct mlx4_spec_list spec_eth_outer = { {NULL} };
 477         struct mlx4_spec_list spec_vxlan     = { {NULL} };
 478         struct mlx4_spec_list spec_eth_inner = { {NULL} };
 479
 480         struct mlx4_net_trans_rule rule = {
 481                 .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 482                 .exclusive = 0,
 483                 .allow_loopback = 1,
 484                 .promisc_mode = MLX4_FS_REGULAR,
 485                 .priority = MLX4_DOMAIN_NIC,
 486         };
 487
 488         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 489
 490         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
 491                 return 0; /* do nothing */
 492
 493         rule.port = priv->port;
 494         rule.qpn = qpn;
 495         INIT_LIST_HEAD(&rule.list);
 496
 497         spec_eth_outer.id = MLX4_NET_TRANS_RULE_ID_ETH;
 498         memcpy(spec_eth_outer.eth.dst_mac, addr, ETH_ALEN);
 499         memcpy(spec_eth_outer.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 500
 501         spec_vxlan.id = MLX4_NET_TRANS_RULE_ID_VXLAN;    /* any vxlan header */
 502         spec_eth_inner.id = MLX4_NET_TRANS_RULE_ID_ETH;  /* any inner eth header */
 503
 504         list_add_tail(&spec_eth_outer.list, &rule.list);
 505         list_add_tail(&spec_vxlan.list,     &rule.list);
 506         list_add_tail(&spec_eth_inner.list, &rule.list);
 507
 508         err = mlx4_flow_attach(priv->mdev->dev, &rule, reg_id);
 509         if (err) {
 510                 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
 511                 return err;
 512         }
 513         en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
 514         return 0;
 515 }
 516
 517
 518 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
 519                                 unsigned char *mac, int *qpn, u64 *reg_id)
 520 {
 521         struct mlx4_en_dev *mdev = priv->mdev;
 522         struct mlx4_dev *dev = mdev->dev;
 523         int err;
 524
 525         switch (dev->caps.steering_mode) {
 526         case MLX4_STEERING_MODE_B0: {
 527                 struct mlx4_qp qp;
 528                 u8 gid[16] = {0};
 529
 530                 qp.qpn = *qpn;
 531                 memcpy(&gid[10], mac, ETH_ALEN);
 532                 gid[5] = priv->port;
 533
 534                 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
 535                 break;
 536         }
 537         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 538                 struct mlx4_spec_list spec_eth = { {NULL} };
 539                 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 540
 541                 struct mlx4_net_trans_rule rule = {
 542                         .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 543                         .exclusive = 0,
 544                         .allow_loopback = 1,
 545                         .promisc_mode = MLX4_FS_REGULAR,
 546                         .priority = MLX4_DOMAIN_NIC,
 547                 };
 548
 549                 rule.port = priv->port;
 550                 rule.qpn = *qpn;
 551                 INIT_LIST_HEAD(&rule.list);
 552
 553                 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
 554                 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
 555                 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 556                 list_add_tail(&spec_eth.list, &rule.list);
 557
 558                 err = mlx4_flow_attach(dev, &rule, reg_id);
 559                 break;
 560         }
 561         default:
 562                 return -EINVAL;
 563         }
 564         if (err)
 565                 en_warn(priv, "Failed Attaching Unicast\n");
 566
 567         return err;
 568 }
 569
 570 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
 571                                      unsigned char *mac, int qpn, u64 reg_id)
 572 {
 573         struct mlx4_en_dev *mdev = priv->mdev;
 574         struct mlx4_dev *dev = mdev->dev;
 575
 576         switch (dev->caps.steering_mode) {
 577         case MLX4_STEERING_MODE_B0: {
 578                 struct mlx4_qp qp;
 579                 u8 gid[16] = {0};
 580
 581                 qp.qpn = qpn;
 582                 memcpy(&gid[10], mac, ETH_ALEN);
 583                 gid[5] = priv->port;
 584
 585                 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
 586                 break;
 587         }
 588         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 589                 mlx4_flow_detach(dev, reg_id);
 590                 break;
 591         }
 592         default:
 593                 en_err(priv, "Invalid steering mode.\n");
 594         }
 595 }
 596
 597 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
 598 {
 599         struct mlx4_en_dev *mdev = priv->mdev;
 600         struct mlx4_dev *dev = mdev->dev;
 601         struct mlx4_mac_entry *entry;
 602         int index = 0;
 603         int err = 0;
 604         u64 reg_id;
 605         int *qpn = &priv->base_qpn;
 606         u64 mac = mlx4_en_mac_to_u64(priv->dev->dev_addr);
 607
 608         en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
 609                priv->dev->dev_addr);
 610         index = mlx4_register_mac(dev, priv->port, mac);
 611         if (index < 0) {
 612                 err = index;
 613                 en_err(priv, "Failed adding MAC: %pM\n",
 614                        priv->dev->dev_addr);
 615                 return err;
 616         }
 617
 618         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 619                 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
 620                 *qpn = base_qpn + index;
 621                 return 0;
 622         }
 623
 624         err = mlx4_qp_reserve_range(dev, 1, 1, qpn);
 625         en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
 626         if (err) {
 627                 en_err(priv, "Failed to reserve qp for mac registration\n");
 628                 goto qp_err;
 629         }
 630
 631         err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
 632         if (err)
 633                 goto steer_err;
 634
 635         if (mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
 636                                      &priv->tunnel_reg_id))
 637                 goto tunnel_err;
 638
 639         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 640         if (!entry) {
 641                 err = -ENOMEM;
 642                 goto alloc_err;
 643         }
 644         memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
 645         entry->reg_id = reg_id;
 646
 647         hlist_add_head_rcu(&entry->hlist,
 648                            &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
 649
 650         return 0;
 651
 652 alloc_err:
 653         if (priv->tunnel_reg_id)
 654                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 655 tunnel_err:
 656         mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
 657
 658 steer_err:
 659         mlx4_qp_release_range(dev, *qpn, 1);
 660
 661 qp_err:
 662         mlx4_unregister_mac(dev, priv->port, mac);
 663         return err;
 664 }
 665
 666 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
 667 {
 668         struct mlx4_en_dev *mdev = priv->mdev;
 669         struct mlx4_dev *dev = mdev->dev;
 670         int qpn = priv->base_qpn;
 671         u64 mac;
 672
 673         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 674                 mac = mlx4_en_mac_to_u64(priv->dev->dev_addr);
 675                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 676                        priv->dev->dev_addr);
 677                 mlx4_unregister_mac(dev, priv->port, mac);
 678         } else {
 679                 struct mlx4_mac_entry *entry;
 680                 struct hlist_node *tmp;
 681                 struct hlist_head *bucket;
 682                 unsigned int i;
 683
 684                 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
 685                         bucket = &priv->mac_hash[i];
 686                         hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 687                                 mac = mlx4_en_mac_to_u64(entry->mac);
 688                                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 689                                        entry->mac);
 690                                 mlx4_en_uc_steer_release(priv, entry->mac,
 691                                                          qpn, entry->reg_id);
 692
 693                                 mlx4_unregister_mac(dev, priv->port, mac);
 694                                 hlist_del_rcu(&entry->hlist);
 695                                 kfree_rcu(entry, rcu);
 696                         }
 697                 }
 698
 699                 if (priv->tunnel_reg_id) {
 700                         mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 701                         priv->tunnel_reg_id = 0;
 702                 }
 703
 704                 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
 705                        priv->port, qpn);
 706                 mlx4_qp_release_range(dev, qpn, 1);
 707                 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
 708         }
 709 }
 710
 711 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
 712                                unsigned char *new_mac, unsigned char *prev_mac)
 713 {
 714         struct mlx4_en_dev *mdev = priv->mdev;
 715         struct mlx4_dev *dev = mdev->dev;
 716         int err = 0;
 717         u64 new_mac_u64 = mlx4_en_mac_to_u64(new_mac);
 718
 719         if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
 720                 struct hlist_head *bucket;
 721                 unsigned int mac_hash;
 722                 struct mlx4_mac_entry *entry;
 723                 struct hlist_node *tmp;
 724                 u64 prev_mac_u64 = mlx4_en_mac_to_u64(prev_mac);
 725
 726                 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
 727                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 728                         if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
 729                                 mlx4_en_uc_steer_release(priv, entry->mac,
 730                                                          qpn, entry->reg_id);
 731                                 mlx4_unregister_mac(dev, priv->port,
 732                                                     prev_mac_u64);
 733                                 hlist_del_rcu(&entry->hlist);
 734                                 synchronize_rcu();
 735                                 memcpy(entry->mac, new_mac, ETH_ALEN);
 736                                 entry->reg_id = 0;
 737                                 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
 738                                 hlist_add_head_rcu(&entry->hlist,
 739                                                    &priv->mac_hash[mac_hash]);
 740                                 mlx4_register_mac(dev, priv->port, new_mac_u64);
 741                                 err = mlx4_en_uc_steer_add(priv, new_mac,
 742                                                            &qpn,
 743                                                            &entry->reg_id);
 744                                 return err;
 745                         }
 746                 }
 747                 return -EINVAL;
 748         }
 749
 750         return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
 751 }
 752
 753 u64 mlx4_en_mac_to_u64(u8 *addr)
 754 {
 755         u64 mac = 0;
 756         int i;
 757
 758         for (i = 0; i < ETH_ALEN; i++) {
 759                 mac <<= 8;
 760                 mac |= addr[i];
 761         }
 762         return mac;
 763 }
 764
 765 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv)
 766 {
 767         int err = 0;
 768
 769         if (priv->port_up) {
 770                 /* Remove old MAC and insert the new one */
 771                 err = mlx4_en_replace_mac(priv, priv->base_qpn,
 772                                           priv->dev->dev_addr, priv->prev_mac);
 773                 if (err)
 774                         en_err(priv, "Failed changing HW MAC address\n");
 775                 memcpy(priv->prev_mac, priv->dev->dev_addr,
 776                        sizeof(priv->prev_mac));
 777         } else
 778                 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
 779
 780         return err;
 781 }
 782
 783 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
 784 {
 785         struct mlx4_en_priv *priv = netdev_priv(dev);
 786         struct mlx4_en_dev *mdev = priv->mdev;
 787         struct sockaddr *saddr = addr;
 788         int err;
 789
 790         if (!is_valid_ether_addr(saddr->sa_data))
 791                 return -EADDRNOTAVAIL;
 792
 793         memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
 794
 795         mutex_lock(&mdev->state_lock);
 796         err = mlx4_en_do_set_mac(priv);
 797         mutex_unlock(&mdev->state_lock);
 798
 799         return err;
 800 }
 801
 802 static void mlx4_en_clear_list(struct net_device *dev)
 803 {
 804         struct mlx4_en_priv *priv = netdev_priv(dev);
 805         struct mlx4_en_mc_list *tmp, *mc_to_del;
 806
 807         list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
 808                 list_del(&mc_to_del->list);
 809                 kfree(mc_to_del);
 810         }
 811 }
 812
 813 static void mlx4_en_cache_mclist(struct net_device *dev)
 814 {
 815         struct mlx4_en_priv *priv = netdev_priv(dev);
 816         struct netdev_hw_addr *ha;
 817         struct mlx4_en_mc_list *tmp;
 818
 819         mlx4_en_clear_list(dev);
 820         netdev_for_each_mc_addr(ha, dev) {
 821                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 822                 if (!tmp) {
 823                         mlx4_en_clear_list(dev);
 824                         return;
 825                 }
 826                 memcpy(tmp->addr, ha->addr, ETH_ALEN);
 827                 list_add_tail(&tmp->list, &priv->mc_list);
 828         }
 829 }
 830
 831 static void update_mclist_flags(struct mlx4_en_priv *priv,
 832                                 struct list_head *dst,
 833                                 struct list_head *src)
 834 {
 835         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 836         bool found;
 837
 838         /* Find all the entries that should be removed from dst,
 839          * These are the entries that are not found in src
 840          */
 841         list_for_each_entry(dst_tmp, dst, list) {
 842                 found = false;
 843                 list_for_each_entry(src_tmp, src, list) {
 844                         if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
 845                                 found = true;
 846                                 break;
 847                         }
 848                 }
 849                 if (!found)
 850                         dst_tmp->action = MCLIST_REM;
 851         }
 852
 853         /* Add entries that exist in src but not in dst
 854          * mark them as need to add
 855          */
 856         list_for_each_entry(src_tmp, src, list) {
 857                 found = false;
 858                 list_for_each_entry(dst_tmp, dst, list) {
 859                         if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
 860                                 dst_tmp->action = MCLIST_NONE;
 861                                 found = true;
 862                                 break;
 863                         }
 864                 }
 865                 if (!found) {
 866                         new_mc = kmemdup(src_tmp,
 867                                          sizeof(struct mlx4_en_mc_list),
 868                                          GFP_KERNEL);
 869                         if (!new_mc)
 870                                 return;
 871
 872                         new_mc->action = MCLIST_ADD;
 873                         list_add_tail(&new_mc->list, dst);
 874                 }
 875         }
 876 }
 877
 878 static void mlx4_en_set_rx_mode(struct net_device *dev)
 879 {
 880         struct mlx4_en_priv *priv = netdev_priv(dev);
 881
 882         if (!priv->port_up)
 883                 return;
 884
 885         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 886 }
 887
 888 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 889                                      struct mlx4_en_dev *mdev)
 890 {
 891         int err = 0;
 892
 893         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 894                 if (netif_msg_rx_status(priv))
 895                         en_warn(priv, "Entering promiscuous mode\n");
 896                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 897
 898                 /* Enable promiscouos mode */
 899                 switch (mdev->dev->caps.steering_mode) {
 900                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 901                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 902                                                           priv->port,
 903                                                           priv->base_qpn,
 904                                                           MLX4_FS_ALL_DEFAULT);
 905                         if (err)
 906                                 en_err(priv, "Failed enabling promiscuous mode\n");
 907                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 908                         break;
 909
 910                 case MLX4_STEERING_MODE_B0:
 911                         err = mlx4_unicast_promisc_add(mdev->dev,
 912                                                        priv->base_qpn,
 913                                                        priv->port);
 914                         if (err)
 915                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 916
 917                         /* Add the default qp number as multicast
 918                          * promisc
 919                          */
 920                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 921                                 err = mlx4_multicast_promisc_add(mdev->dev,
 922                                                                  priv->base_qpn,
 923                                                                  priv->port);
 924                                 if (err)
 925                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 926                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 927                         }
 928                         break;
 929
 930                 case MLX4_STEERING_MODE_A0:
 931                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 932                                                      priv->port,
 933                                                      priv->base_qpn,
 934                                                      1);
 935                         if (err)
 936                                 en_err(priv, "Failed enabling promiscuous mode\n");
 937                         break;
 938                 }
 939
 940                 /* Disable port multicast filter (unconditionally) */
 941                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 942                                           0, MLX4_MCAST_DISABLE);
 943                 if (err)
 944                         en_err(priv, "Failed disabling multicast filter\n");
 945
 946                 /* Disable port VLAN filter */
 947                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 948                 if (err)
 949                         en_err(priv, "Failed disabling VLAN filter\n");
 950         }
 951 }
 952
 953 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 954                                        struct mlx4_en_dev *mdev)
 955 {
 956         int err = 0;
 957
 958         if (netif_msg_rx_status(priv))
 959                 en_warn(priv, "Leaving promiscuous mode\n");
 960         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 961
 962         /* Disable promiscouos mode */
 963         switch (mdev->dev->caps.steering_mode) {
 964         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 965                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 966                                                      priv->port,
 967                                                      MLX4_FS_ALL_DEFAULT);
 968                 if (err)
 969                         en_err(priv, "Failed disabling promiscuous mode\n");
 970                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 971                 break;
 972
 973         case MLX4_STEERING_MODE_B0:
 974                 err = mlx4_unicast_promisc_remove(mdev->dev,
 975                                                   priv->base_qpn,
 976                                                   priv->port);
 977                 if (err)
 978                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 979                 /* Disable Multicast promisc */
 980                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 981                         err = mlx4_multicast_promisc_remove(mdev->dev,
 982                                                             priv->base_qpn,
 983                                                             priv->port);
 984                         if (err)
 985                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 986                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 987                 }
 988                 break;
 989
 990         case MLX4_STEERING_MODE_A0:
 991                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 992                                              priv->port,
 993                                              priv->base_qpn, 0);
 994                 if (err)
 995                         en_err(priv, "Failed disabling promiscuous mode\n");
 996                 break;
 997         }
 998
 999         /* Enable port VLAN filter */
1000         err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
1001         if (err)
1002                 en_err(priv, "Failed enabling VLAN filter\n");
1003 }
1004
1005 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
1006                                  struct net_device *dev,
1007                                  struct mlx4_en_dev *mdev)
1008 {
1009         struct mlx4_en_mc_list *mclist, *tmp;
1010         u64 mcast_addr = 0;
1011         u8 mc_list[16] = {0};
1012         int err = 0;
1013
1014         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
1015         if (dev->flags & IFF_ALLMULTI) {
1016                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1017                                           0, MLX4_MCAST_DISABLE);
1018                 if (err)
1019                         en_err(priv, "Failed disabling multicast filter\n");
1020
1021                 /* Add the default qp number as multicast promisc */
1022                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
1023                         switch (mdev->dev->caps.steering_mode) {
1024                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
1025                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
1026                                                                   priv->port,
1027                                                                   priv->base_qpn,
1028                                                                   MLX4_FS_MC_DEFAULT);
1029                                 break;
1030
1031                         case MLX4_STEERING_MODE_B0:
1032                                 err = mlx4_multicast_promisc_add(mdev->dev,
1033                                                                  priv->base_qpn,
1034                                                                  priv->port);
1035                                 break;
1036
1037                         case MLX4_STEERING_MODE_A0:
1038                                 break;
1039                         }
1040                         if (err)
1041                                 en_err(priv, "Failed entering multicast promisc mode\n");
1042                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1043                 }
1044         } else {
1045                 /* Disable Multicast promisc */
1046                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1047                         switch (mdev->dev->caps.steering_mode) {
1048                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
1049                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
1050                                                                      priv->port,
1051                                                                      MLX4_FS_MC_DEFAULT);
1052                                 break;
1053
1054                         case MLX4_STEERING_MODE_B0:
1055                                 err = mlx4_multicast_promisc_remove(mdev->dev,
1056                                                                     priv->base_qpn,
1057                                                                     priv->port);
1058                                 break;
1059
1060                         case MLX4_STEERING_MODE_A0:
1061                                 break;
1062                         }
1063                         if (err)
1064                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
1065                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1066                 }
1067
1068                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1069                                           0, MLX4_MCAST_DISABLE);
1070                 if (err)
1071                         en_err(priv, "Failed disabling multicast filter\n");
1072
1073                 /* Flush mcast filter and init it with broadcast address */
1074                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1075                                     1, MLX4_MCAST_CONFIG);
1076
1077                 /* Update multicast list - we cache all addresses so they won't
1078                  * change while HW is updated holding the command semaphor */
1079                 netif_addr_lock_bh(dev);
1080                 mlx4_en_cache_mclist(dev);
1081                 netif_addr_unlock_bh(dev);
1082                 list_for_each_entry(mclist, &priv->mc_list, list) {
1083                         mcast_addr = mlx4_en_mac_to_u64(mclist->addr);
1084                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1085                                             mcast_addr, 0, MLX4_MCAST_CONFIG);
1086                 }
1087                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1088                                           0, MLX4_MCAST_ENABLE);
1089                 if (err)
1090                         en_err(priv, "Failed enabling multicast filter\n");
1091
1092                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1093                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1094                         if (mclist->action == MCLIST_REM) {
1095                                 /* detach this address and delete from list */
1096                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1097                                 mc_list[5] = priv->port;
1098                                 err = mlx4_multicast_detach(mdev->dev,
1099                                                             &priv->rss_map.indir_qp,
1100                                                             mc_list,
1101                                                             MLX4_PROT_ETH,
1102                                                             mclist->reg_id);
1103                                 if (err)
1104                                         en_err(priv, "Fail to detach multicast address\n");
1105
1106                                 if (mclist->tunnel_reg_id) {
1107                                         err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1108                                         if (err)
1109                                                 en_err(priv, "Failed to detach multicast address\n");
1110                                 }
1111
1112                                 /* remove from list */
1113                                 list_del(&mclist->list);
1114                                 kfree(mclist);
1115                         } else if (mclist->action == MCLIST_ADD) {
1116                                 /* attach the address */
1117                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1118                                 /* needed for B0 steering support */
1119                                 mc_list[5] = priv->port;
1120                                 err = mlx4_multicast_attach(mdev->dev,
1121                                                             &priv->rss_map.indir_qp,
1122                                                             mc_list,
1123                                                             priv->port, 0,
1124                                                             MLX4_PROT_ETH,
1125                                                             &mclist->reg_id);
1126                                 if (err)
1127                                         en_err(priv, "Fail to attach multicast address\n");
1128
1129                                 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1130                                                                &mclist->tunnel_reg_id);
1131                                 if (err)
1132                                         en_err(priv, "Failed to attach multicast address\n");
1133                         }
1134                 }
1135         }
1136 }
1137
1138 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1139                                  struct net_device *dev,
1140                                  struct mlx4_en_dev *mdev)
1141 {
1142         struct netdev_hw_addr *ha;
1143         struct mlx4_mac_entry *entry;
1144         struct hlist_node *tmp;
1145         bool found;
1146         u64 mac;
1147         int err = 0;
1148         struct hlist_head *bucket;
1149         unsigned int i;
1150         int removed = 0;
1151         u32 prev_flags;
1152
1153         /* Note that we do not need to protect our mac_hash traversal with rcu,
1154          * since all modification code is protected by mdev->state_lock
1155          */
1156
1157         /* find what to remove */
1158         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1159                 bucket = &priv->mac_hash[i];
1160                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1161                         found = false;
1162                         netdev_for_each_uc_addr(ha, dev) {
1163                                 if (ether_addr_equal_64bits(entry->mac,
1164                                                             ha->addr)) {
1165                                         found = true;
1166                                         break;
1167                                 }
1168                         }
1169
1170                         /* MAC address of the port is not in uc list */
1171                         if (ether_addr_equal_64bits(entry->mac, dev->dev_addr))
1172                                 found = true;
1173
1174                         if (!found) {
1175                                 mac = mlx4_en_mac_to_u64(entry->mac);
1176                                 mlx4_en_uc_steer_release(priv, entry->mac,
1177                                                          priv->base_qpn,
1178                                                          entry->reg_id);
1179                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1180
1181                                 hlist_del_rcu(&entry->hlist);
1182                                 kfree_rcu(entry, rcu);
1183                                 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1184                                        entry->mac, priv->port);
1185                                 ++removed;
1186                         }
1187                 }
1188         }
1189
1190         /* if we didn't remove anything, there is no use in trying to add
1191          * again once we are in a forced promisc mode state
1192          */
1193         if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1194                 return;
1195
1196         prev_flags = priv->flags;
1197         priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1198
1199         /* find what to add */
1200         netdev_for_each_uc_addr(ha, dev) {
1201                 found = false;
1202                 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1203                 hlist_for_each_entry(entry, bucket, hlist) {
1204                         if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1205                                 found = true;
1206                                 break;
1207                         }
1208                 }
1209
1210                 if (!found) {
1211                         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1212                         if (!entry) {
1213                                 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1214                                        ha->addr, priv->port);
1215                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1216                                 break;
1217                         }
1218                         mac = mlx4_en_mac_to_u64(ha->addr);
1219                         memcpy(entry->mac, ha->addr, ETH_ALEN);
1220                         err = mlx4_register_mac(mdev->dev, priv->port, mac);
1221                         if (err < 0) {
1222                                 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1223                                        ha->addr, priv->port, err);
1224                                 kfree(entry);
1225                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1226                                 break;
1227                         }
1228                         err = mlx4_en_uc_steer_add(priv, ha->addr,
1229                                                    &priv->base_qpn,
1230                                                    &entry->reg_id);
1231                         if (err) {
1232                                 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1233                                        ha->addr, priv->port, err);
1234                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1235                                 kfree(entry);
1236                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1237                                 break;
1238                         } else {
1239                                 unsigned int mac_hash;
1240                                 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1241                                        ha->addr, priv->port);
1242                                 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1243                                 bucket = &priv->mac_hash[mac_hash];
1244                                 hlist_add_head_rcu(&entry->hlist, bucket);
1245                         }
1246                 }
1247         }
1248
1249         if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1250                 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1251                         priv->port);
1252         } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1253                 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1254                         priv->port);
1255         }
1256 }
1257
1258 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1259 {
1260         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1261                                                  rx_mode_task);
1262         struct mlx4_en_dev *mdev = priv->mdev;
1263         struct net_device *dev = priv->dev;
1264
1265         mutex_lock(&mdev->state_lock);
1266         if (!mdev->device_up) {
1267                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1268                 goto out;
1269         }
1270         if (!priv->port_up) {
1271                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1272                 goto out;
1273         }
1274
1275         if (!netif_carrier_ok(dev)) {
1276                 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1277                         if (priv->port_state.link_state) {
1278                                 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1279                                 netif_carrier_on(dev);
1280                                 en_dbg(LINK, priv, "Link Up\n");
1281                         }
1282                 }
1283         }
1284
1285         if (dev->priv_flags & IFF_UNICAST_FLT)
1286                 mlx4_en_do_uc_filter(priv, dev, mdev);
1287
1288         /* Promsicuous mode: disable all filters */
1289         if ((dev->flags & IFF_PROMISC) ||
1290             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1291                 mlx4_en_set_promisc_mode(priv, mdev);
1292                 goto out;
1293         }
1294
1295         /* Not in promiscuous mode */
1296         if (priv->flags & MLX4_EN_FLAG_PROMISC)
1297                 mlx4_en_clear_promisc_mode(priv, mdev);
1298
1299         mlx4_en_do_multicast(priv, dev, mdev);
1300 out:
1301         mutex_unlock(&mdev->state_lock);
1302 }
1303
1304 #ifdef CONFIG_NET_POLL_CONTROLLER
1305 static void mlx4_en_netpoll(struct net_device *dev)
1306 {
1307         struct mlx4_en_priv *priv = netdev_priv(dev);
1308         struct mlx4_en_cq *cq;
1309         unsigned long flags;
1310         int i;
1311
1312         for (i = 0; i < priv->rx_ring_num; i++) {
1313                 cq = priv->rx_cq[i];
1314                 spin_lock_irqsave(&cq->lock, flags);
1315                 napi_synchronize(&cq->napi);
1316                 mlx4_en_process_rx_cq(dev, cq, 0);
1317                 spin_unlock_irqrestore(&cq->lock, flags);
1318         }
1319 }
1320 #endif
1321
1322 static void mlx4_en_tx_timeout(struct net_device *dev)
1323 {
1324         struct mlx4_en_priv *priv = netdev_priv(dev);
1325         struct mlx4_en_dev *mdev = priv->mdev;
1326         int i;
1327
1328         if (netif_msg_timer(priv))
1329                 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1330
1331         for (i = 0; i < priv->tx_ring_num; i++) {
1332                 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1333                         continue;
1334                 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1335                         i, priv->tx_ring[i]->qpn, priv->tx_ring[i]->cqn,
1336                         priv->tx_ring[i]->cons, priv->tx_ring[i]->prod);
1337         }
1338
1339         priv->port_stats.tx_timeout++;
1340         en_dbg(DRV, priv, "Scheduling watchdog\n");
1341         queue_work(mdev->workqueue, &priv->watchdog_task);
1342 }
1343
1344
1345 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
1346 {
1347         struct mlx4_en_priv *priv = netdev_priv(dev);
1348
1349         spin_lock_bh(&priv->stats_lock);
1350         memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
1351         spin_unlock_bh(&priv->stats_lock);
1352
1353         return &priv->ret_stats;
1354 }
1355
1356 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1357 {
1358         struct mlx4_en_cq *cq;
1359         int i;
1360
1361         /* If we haven't received a specific coalescing setting
1362          * (module param), we set the moderation parameters as follows:
1363          * - moder_cnt is set to the number of mtu sized packets to
1364          *   satisfy our coalescing target.
1365          * - moder_time is set to a fixed value.
1366          */
1367         priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1368         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1369         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1370         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1371         en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1372                priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1373
1374         /* Setup cq moderation params */
1375         for (i = 0; i < priv->rx_ring_num; i++) {
1376                 cq = priv->rx_cq[i];
1377                 cq->moder_cnt = priv->rx_frames;
1378                 cq->moder_time = priv->rx_usecs;
1379                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1380                 priv->last_moder_packets[i] = 0;
1381                 priv->last_moder_bytes[i] = 0;
1382         }
1383
1384         for (i = 0; i < priv->tx_ring_num; i++) {
1385                 cq = priv->tx_cq[i];
1386                 cq->moder_cnt = priv->tx_frames;
1387                 cq->moder_time = priv->tx_usecs;
1388         }
1389
1390         /* Reset auto-moderation params */
1391         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1392         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1393         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1394         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1395         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1396         priv->adaptive_rx_coal = 1;
1397         priv->last_moder_jiffies = 0;
1398         priv->last_moder_tx_packets = 0;
1399 }
1400
1401 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1402 {
1403         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1404         struct mlx4_en_cq *cq;
1405         unsigned long packets;
1406         unsigned long rate;
1407         unsigned long avg_pkt_size;
1408         unsigned long rx_packets;
1409         unsigned long rx_bytes;
1410         unsigned long rx_pkt_diff;
1411         int moder_time;
1412         int ring, err;
1413
1414         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1415                 return;
1416
1417         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1418                 spin_lock_bh(&priv->stats_lock);
1419                 rx_packets = priv->rx_ring[ring]->packets;
1420                 rx_bytes = priv->rx_ring[ring]->bytes;
1421                 spin_unlock_bh(&priv->stats_lock);
1422
1423                 rx_pkt_diff = ((unsigned long) (rx_packets -
1424                                 priv->last_moder_packets[ring]));
1425                 packets = rx_pkt_diff;
1426                 rate = packets * HZ / period;
1427                 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1428                                 priv->last_moder_bytes[ring])) / packets : 0;
1429
1430                 /* Apply auto-moderation only when packet rate
1431                  * exceeds a rate that it matters */
1432                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1433                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1434                         if (rate < priv->pkt_rate_low)
1435                                 moder_time = priv->rx_usecs_low;
1436                         else if (rate > priv->pkt_rate_high)
1437                                 moder_time = priv->rx_usecs_high;
1438                         else
1439                                 moder_time = (rate - priv->pkt_rate_low) *
1440                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1441                                         (priv->pkt_rate_high - priv->pkt_rate_low) +
1442                                         priv->rx_usecs_low;
1443                 } else {
1444                         moder_time = priv->rx_usecs_low;
1445                 }
1446
1447                 if (moder_time != priv->last_moder_time[ring]) {
1448                         priv->last_moder_time[ring] = moder_time;
1449                         cq = priv->rx_cq[ring];
1450                         cq->moder_time = moder_time;
1451                         cq->moder_cnt = priv->rx_frames;
1452                         err = mlx4_en_set_cq_moder(priv, cq);
1453                         if (err)
1454                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1455                                        ring);
1456                 }
1457                 priv->last_moder_packets[ring] = rx_packets;
1458                 priv->last_moder_bytes[ring] = rx_bytes;
1459         }
1460
1461         priv->last_moder_jiffies = jiffies;
1462 }
1463
1464 static void mlx4_en_do_get_stats(struct work_struct *work)
1465 {
1466         struct delayed_work *delay = to_delayed_work(work);
1467         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1468                                                  stats_task);
1469         struct mlx4_en_dev *mdev = priv->mdev;
1470         int err;
1471
1472         mutex_lock(&mdev->state_lock);
1473         if (mdev->device_up) {
1474                 if (priv->port_up) {
1475                         err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1476                         if (err)
1477                                 en_dbg(HW, priv, "Could not update stats\n");
1478
1479                         mlx4_en_auto_moderation(priv);
1480                 }
1481
1482                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1483         }
1484         if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1485                 mlx4_en_do_set_mac(priv);
1486                 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1487         }
1488         mutex_unlock(&mdev->state_lock);
1489 }
1490
1491 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1492  * periodically
1493  */
1494 static void mlx4_en_service_task(struct work_struct *work)
1495 {
1496         struct delayed_work *delay = to_delayed_work(work);
1497         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1498                                                  service_task);
1499         struct mlx4_en_dev *mdev = priv->mdev;
1500
1501         mutex_lock(&mdev->state_lock);
1502         if (mdev->device_up) {
1503                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1504                         mlx4_en_ptp_overflow_check(mdev);
1505
1506                 queue_delayed_work(mdev->workqueue, &priv->service_task,
1507                                    SERVICE_TASK_DELAY);
1508         }
1509         mutex_unlock(&mdev->state_lock);
1510 }
1511
1512 static void mlx4_en_linkstate(struct work_struct *work)
1513 {
1514         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1515                                                  linkstate_task);
1516         struct mlx4_en_dev *mdev = priv->mdev;
1517         int linkstate = priv->link_state;
1518
1519         mutex_lock(&mdev->state_lock);
1520         /* If observable port state changed set carrier state and
1521          * report to system log */
1522         if (priv->last_link_state != linkstate) {
1523                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1524                         en_info(priv, "Link Down\n");
1525                         netif_carrier_off(priv->dev);
1526                 } else {
1527                         en_info(priv, "Link Up\n");
1528                         netif_carrier_on(priv->dev);
1529                 }
1530         }
1531         priv->last_link_state = linkstate;
1532         mutex_unlock(&mdev->state_lock);
1533 }
1534
1535
1536 int mlx4_en_start_port(struct net_device *dev)
1537 {
1538         struct mlx4_en_priv *priv = netdev_priv(dev);
1539         struct mlx4_en_dev *mdev = priv->mdev;
1540         struct mlx4_en_cq *cq;
1541         struct mlx4_en_tx_ring *tx_ring;
1542         int rx_index = 0;
1543         int tx_index = 0;
1544         int err = 0;
1545         int i;
1546         int j;
1547         u8 mc_list[16] = {0};
1548
1549         if (priv->port_up) {
1550                 en_dbg(DRV, priv, "start port called while port already up\n");
1551                 return 0;
1552         }
1553
1554         INIT_LIST_HEAD(&priv->mc_list);
1555         INIT_LIST_HEAD(&priv->curr_list);
1556         INIT_LIST_HEAD(&priv->ethtool_list);
1557         memset(&priv->ethtool_rules[0], 0,
1558                sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1559
1560         /* Calculate Rx buf size */
1561         dev->mtu = min(dev->mtu, priv->max_mtu);
1562         mlx4_en_calc_rx_buf(dev);
1563         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1564
1565         /* Configure rx cq's and rings */
1566         err = mlx4_en_activate_rx_rings(priv);
1567         if (err) {
1568                 en_err(priv, "Failed to activate RX rings\n");
1569                 return err;
1570         }
1571         for (i = 0; i < priv->rx_ring_num; i++) {
1572                 cq = priv->rx_cq[i];
1573
1574                 mlx4_en_cq_init_lock(cq);
1575
1576                 err = mlx4_en_activate_cq(priv, cq, i);
1577                 if (err) {
1578                         en_err(priv, "Failed activating Rx CQ\n");
1579                         goto cq_err;
1580                 }
1581                 for (j = 0; j < cq->size; j++)
1582                         cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1583                 err = mlx4_en_set_cq_moder(priv, cq);
1584                 if (err) {
1585                         en_err(priv, "Failed setting cq moderation parameters");
1586                         mlx4_en_deactivate_cq(priv, cq);
1587                         goto cq_err;
1588                 }
1589                 mlx4_en_arm_cq(priv, cq);
1590                 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1591                 ++rx_index;
1592         }
1593
1594         /* Set qp number */
1595         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1596         err = mlx4_en_get_qp(priv);
1597         if (err) {
1598                 en_err(priv, "Failed getting eth qp\n");
1599                 goto cq_err;
1600         }
1601         mdev->mac_removed[priv->port] = 0;
1602
1603         err = mlx4_en_config_rss_steer(priv);
1604         if (err) {
1605                 en_err(priv, "Failed configuring rss steering\n");
1606                 goto mac_err;
1607         }
1608
1609         err = mlx4_en_create_drop_qp(priv);
1610         if (err)
1611                 goto rss_err;
1612
1613         /* Configure tx cq's and rings */
1614         for (i = 0; i < priv->tx_ring_num; i++) {
1615                 /* Configure cq */
1616                 cq = priv->tx_cq[i];
1617                 err = mlx4_en_activate_cq(priv, cq, i);
1618                 if (err) {
1619                         en_err(priv, "Failed allocating Tx CQ\n");
1620                         goto tx_err;
1621                 }
1622                 err = mlx4_en_set_cq_moder(priv, cq);
1623                 if (err) {
1624                         en_err(priv, "Failed setting cq moderation parameters");
1625                         mlx4_en_deactivate_cq(priv, cq);
1626                         goto tx_err;
1627                 }
1628                 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1629                 cq->buf->wqe_index = cpu_to_be16(0xffff);
1630
1631                 /* Configure ring */
1632                 tx_ring = priv->tx_ring[i];
1633                 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1634                         i / priv->num_tx_rings_p_up);
1635                 if (err) {
1636                         en_err(priv, "Failed allocating Tx ring\n");
1637                         mlx4_en_deactivate_cq(priv, cq);
1638                         goto tx_err;
1639                 }
1640                 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1641
1642                 /* Arm CQ for TX completions */
1643                 mlx4_en_arm_cq(priv, cq);
1644
1645                 /* Set initial ownership of all Tx TXBBs to SW (1) */
1646                 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1647                         *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1648                 ++tx_index;
1649         }
1650
1651         /* Configure port */
1652         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1653                                     priv->rx_skb_size + ETH_FCS_LEN,
1654                                     priv->prof->tx_pause,
1655                                     priv->prof->tx_ppp,
1656                                     priv->prof->rx_pause,
1657                                     priv->prof->rx_ppp);
1658         if (err) {
1659                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1660                        priv->port, err);
1661                 goto tx_err;
1662         }
1663         /* Set default qp number */
1664         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1665         if (err) {
1666                 en_err(priv, "Failed setting default qp numbers\n");
1667                 goto tx_err;
1668         }
1669
1670         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1671                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC);
1672                 if (err) {
1673                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1674                                err);
1675                         goto tx_err;
1676                 }
1677         }
1678
1679         /* Init port */
1680         en_dbg(HW, priv, "Initializing port\n");
1681         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1682         if (err) {
1683                 en_err(priv, "Failed Initializing port\n");
1684                 goto tx_err;
1685         }
1686
1687         /* Attach rx QP to bradcast address */
1688         memset(&mc_list[10], 0xff, ETH_ALEN);
1689         mc_list[5] = priv->port; /* needed for B0 steering support */
1690         if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1691                                   priv->port, 0, MLX4_PROT_ETH,
1692                                   &priv->broadcast_id))
1693                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1694
1695         /* Must redo promiscuous mode setup. */
1696         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1697
1698         /* Schedule multicast task to populate multicast list */
1699         queue_work(mdev->workqueue, &priv->rx_mode_task);
1700
1701         mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
1702
1703         priv->port_up = true;
1704         netif_tx_start_all_queues(dev);
1705         netif_device_attach(dev);
1706
1707         return 0;
1708
1709 tx_err:
1710         while (tx_index--) {
1711                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1712                 mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1713         }
1714         mlx4_en_destroy_drop_qp(priv);
1715 rss_err:
1716         mlx4_en_release_rss_steer(priv);
1717 mac_err:
1718         mlx4_en_put_qp(priv);
1719 cq_err:
1720         while (rx_index--)
1721                 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1722         for (i = 0; i < priv->rx_ring_num; i++)
1723                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1724
1725         return err; /* need to close devices */
1726 }
1727
1728
1729 void mlx4_en_stop_port(struct net_device *dev, int detach)
1730 {
1731         struct mlx4_en_priv *priv = netdev_priv(dev);
1732         struct mlx4_en_dev *mdev = priv->mdev;
1733         struct mlx4_en_mc_list *mclist, *tmp;
1734         struct ethtool_flow_id *flow, *tmp_flow;
1735         int i;
1736         u8 mc_list[16] = {0};
1737
1738         if (!priv->port_up) {
1739                 en_dbg(DRV, priv, "stop port called while port already down\n");
1740                 return;
1741         }
1742
1743         /* close port*/
1744         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1745
1746         /* Synchronize with tx routine */
1747         netif_tx_lock_bh(dev);
1748         if (detach)
1749                 netif_device_detach(dev);
1750         netif_tx_stop_all_queues(dev);
1751         netif_tx_unlock_bh(dev);
1752
1753         netif_tx_disable(dev);
1754
1755         /* Set port as not active */
1756         priv->port_up = false;
1757
1758         /* Promsicuous mode */
1759         if (mdev->dev->caps.steering_mode ==
1760             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1761                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1762                                  MLX4_EN_FLAG_MC_PROMISC);
1763                 mlx4_flow_steer_promisc_remove(mdev->dev,
1764                                                priv->port,
1765                                                MLX4_FS_ALL_DEFAULT);
1766                 mlx4_flow_steer_promisc_remove(mdev->dev,
1767                                                priv->port,
1768                                                MLX4_FS_MC_DEFAULT);
1769         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1770                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1771
1772                 /* Disable promiscouos mode */
1773                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1774                                             priv->port);
1775
1776                 /* Disable Multicast promisc */
1777                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1778                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1779                                                       priv->port);
1780                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1781                 }
1782         }
1783
1784         /* Detach All multicasts */
1785         memset(&mc_list[10], 0xff, ETH_ALEN);
1786         mc_list[5] = priv->port; /* needed for B0 steering support */
1787         mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1788                               MLX4_PROT_ETH, priv->broadcast_id);
1789         list_for_each_entry(mclist, &priv->curr_list, list) {
1790                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1791                 mc_list[5] = priv->port;
1792                 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1793                                       mc_list, MLX4_PROT_ETH, mclist->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                 while (test_bit(NAPI_STATE_SCHED, &cq->napi.state))
1847                         msleep(1);
1848                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1849                 mlx4_en_deactivate_cq(priv, cq);
1850         }
1851 }
1852
1853 static void mlx4_en_restart(struct work_struct *work)
1854 {
1855         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1856                                                  watchdog_task);
1857         struct mlx4_en_dev *mdev = priv->mdev;
1858         struct net_device *dev = priv->dev;
1859
1860         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1861
1862         mutex_lock(&mdev->state_lock);
1863         if (priv->port_up) {
1864                 mlx4_en_stop_port(dev, 1);
1865                 if (mlx4_en_start_port(dev))
1866                         en_err(priv, "Failed restarting port %d\n", priv->port);
1867         }
1868         mutex_unlock(&mdev->state_lock);
1869 }
1870
1871 static void mlx4_en_clear_stats(struct net_device *dev)
1872 {
1873         struct mlx4_en_priv *priv = netdev_priv(dev);
1874         struct mlx4_en_dev *mdev = priv->mdev;
1875         int i;
1876
1877         if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1878                 en_dbg(HW, priv, "Failed dumping statistics\n");
1879
1880         memset(&priv->stats, 0, sizeof(priv->stats));
1881         memset(&priv->pstats, 0, sizeof(priv->pstats));
1882         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1883         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1884
1885         for (i = 0; i < priv->tx_ring_num; i++) {
1886                 priv->tx_ring[i]->bytes = 0;
1887                 priv->tx_ring[i]->packets = 0;
1888                 priv->tx_ring[i]->tx_csum = 0;
1889         }
1890         for (i = 0; i < priv->rx_ring_num; i++) {
1891                 priv->rx_ring[i]->bytes = 0;
1892                 priv->rx_ring[i]->packets = 0;
1893                 priv->rx_ring[i]->csum_ok = 0;
1894                 priv->rx_ring[i]->csum_none = 0;
1895         }
1896 }
1897
1898 static int mlx4_en_open(struct net_device *dev)
1899 {
1900         struct mlx4_en_priv *priv = netdev_priv(dev);
1901         struct mlx4_en_dev *mdev = priv->mdev;
1902         int err = 0;
1903
1904         mutex_lock(&mdev->state_lock);
1905
1906         if (!mdev->device_up) {
1907                 en_err(priv, "Cannot open - device down/disabled\n");
1908                 err = -EBUSY;
1909                 goto out;
1910         }
1911
1912         /* Reset HW statistics and SW counters */
1913         mlx4_en_clear_stats(dev);
1914
1915         err = mlx4_en_start_port(dev);
1916         if (err)
1917                 en_err(priv, "Failed starting port:%d\n", priv->port);
1918
1919 out:
1920         mutex_unlock(&mdev->state_lock);
1921         return err;
1922 }
1923
1924
1925 static int mlx4_en_close(struct net_device *dev)
1926 {
1927         struct mlx4_en_priv *priv = netdev_priv(dev);
1928         struct mlx4_en_dev *mdev = priv->mdev;
1929
1930         en_dbg(IFDOWN, priv, "Close port called\n");
1931
1932         mutex_lock(&mdev->state_lock);
1933
1934         mlx4_en_stop_port(dev, 0);
1935         netif_carrier_off(dev);
1936
1937         mutex_unlock(&mdev->state_lock);
1938         return 0;
1939 }
1940
1941 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1942 {
1943         int i;
1944
1945 #ifdef CONFIG_RFS_ACCEL
1946         free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1947         priv->dev->rx_cpu_rmap = NULL;
1948 #endif
1949
1950         for (i = 0; i < priv->tx_ring_num; i++) {
1951                 if (priv->tx_ring && priv->tx_ring[i])
1952                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1953                 if (priv->tx_cq && priv->tx_cq[i])
1954                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1955         }
1956
1957         for (i = 0; i < priv->rx_ring_num; i++) {
1958                 if (priv->rx_ring[i])
1959                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1960                                 priv->prof->rx_ring_size, priv->stride);
1961                 if (priv->rx_cq[i])
1962                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1963         }
1964
1965         if (priv->base_tx_qpn) {
1966                 mlx4_qp_release_range(priv->mdev->dev, priv->base_tx_qpn, priv->tx_ring_num);
1967                 priv->base_tx_qpn = 0;
1968         }
1969 }
1970
1971 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1972 {
1973         struct mlx4_en_port_profile *prof = priv->prof;
1974         int i;
1975         int err;
1976         int node;
1977
1978         err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &priv->base_tx_qpn);
1979         if (err) {
1980                 en_err(priv, "failed reserving range for TX rings\n");
1981                 return err;
1982         }
1983
1984         /* Create tx Rings */
1985         for (i = 0; i < priv->tx_ring_num; i++) {
1986                 node = cpu_to_node(i % num_online_cpus());
1987                 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1988                                       prof->tx_ring_size, i, TX, node))
1989                         goto err;
1990
1991                 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
1992                                            priv->base_tx_qpn + i,
1993                                            prof->tx_ring_size, TXBB_SIZE,
1994                                            node, i))
1995                         goto err;
1996         }
1997
1998         /* Create rx Rings */
1999         for (i = 0; i < priv->rx_ring_num; i++) {
2000                 node = cpu_to_node(i % num_online_cpus());
2001                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2002                                       prof->rx_ring_size, i, RX, node))
2003                         goto err;
2004
2005                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2006                                            prof->rx_ring_size, priv->stride,
2007                                            node))
2008                         goto err;
2009         }
2010
2011 #ifdef CONFIG_RFS_ACCEL
2012         if (priv->mdev->dev->caps.comp_pool) {
2013                 priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->mdev->dev->caps.comp_pool);
2014                 if (!priv->dev->rx_cpu_rmap)
2015                         goto err;
2016         }
2017 #endif
2018
2019         return 0;
2020
2021 err:
2022         en_err(priv, "Failed to allocate NIC resources\n");
2023         for (i = 0; i < priv->rx_ring_num; i++) {
2024                 if (priv->rx_ring[i])
2025                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2026                                                 prof->rx_ring_size,
2027                                                 priv->stride);
2028                 if (priv->rx_cq[i])
2029                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2030         }
2031         for (i = 0; i < priv->tx_ring_num; i++) {
2032                 if (priv->tx_ring[i])
2033                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
2034                 if (priv->tx_cq[i])
2035                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
2036         }
2037         return -ENOMEM;
2038 }
2039
2040
2041 void mlx4_en_destroy_netdev(struct net_device *dev)
2042 {
2043         struct mlx4_en_priv *priv = netdev_priv(dev);
2044         struct mlx4_en_dev *mdev = priv->mdev;
2045
2046         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2047
2048         /* Unregister device - this will close the port if it was up */
2049         if (priv->registered)
2050                 unregister_netdev(dev);
2051
2052         if (priv->allocated)
2053                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2054
2055         cancel_delayed_work(&priv->stats_task);
2056         cancel_delayed_work(&priv->service_task);
2057         /* flush any pending task for this netdev */
2058         flush_workqueue(mdev->workqueue);
2059
2060         /* Detach the netdev so tasks would not attempt to access it */
2061         mutex_lock(&mdev->state_lock);
2062         mdev->pndev[priv->port] = NULL;
2063         mutex_unlock(&mdev->state_lock);
2064
2065         mlx4_en_free_resources(priv);
2066
2067         kfree(priv->tx_ring);
2068         kfree(priv->tx_cq);
2069
2070         free_netdev(dev);
2071 }
2072
2073 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2074 {
2075         struct mlx4_en_priv *priv = netdev_priv(dev);
2076         struct mlx4_en_dev *mdev = priv->mdev;
2077         int err = 0;
2078
2079         en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2080                  dev->mtu, new_mtu);
2081
2082         if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
2083                 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
2084                 return -EPERM;
2085         }
2086         dev->mtu = new_mtu;
2087
2088         if (netif_running(dev)) {
2089                 mutex_lock(&mdev->state_lock);
2090                 if (!mdev->device_up) {
2091                         /* NIC is probably restarting - let watchdog task reset
2092                          * the port */
2093                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2094                 } else {
2095                         mlx4_en_stop_port(dev, 1);
2096                         err = mlx4_en_start_port(dev);
2097                         if (err) {
2098                                 en_err(priv, "Failed restarting port:%d\n",
2099                                          priv->port);
2100                                 queue_work(mdev->workqueue, &priv->watchdog_task);
2101                         }
2102                 }
2103                 mutex_unlock(&mdev->state_lock);
2104         }
2105         return 0;
2106 }
2107
2108 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2109 {
2110         struct mlx4_en_priv *priv = netdev_priv(dev);
2111         struct mlx4_en_dev *mdev = priv->mdev;
2112         struct hwtstamp_config config;
2113
2114         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2115                 return -EFAULT;
2116
2117         /* reserved for future extensions */
2118         if (config.flags)
2119                 return -EINVAL;
2120
2121         /* device doesn't support time stamping */
2122         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2123                 return -EINVAL;
2124
2125         /* TX HW timestamp */
2126         switch (config.tx_type) {
2127         case HWTSTAMP_TX_OFF:
2128         case HWTSTAMP_TX_ON:
2129                 break;
2130         default:
2131                 return -ERANGE;
2132         }
2133
2134         /* RX HW timestamp */
2135         switch (config.rx_filter) {
2136         case HWTSTAMP_FILTER_NONE:
2137                 break;
2138         case HWTSTAMP_FILTER_ALL:
2139         case HWTSTAMP_FILTER_SOME:
2140         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2141         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2142         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2143         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2144         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2145         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2146         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2147         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2148         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2149         case HWTSTAMP_FILTER_PTP_V2_EVENT:
2150         case HWTSTAMP_FILTER_PTP_V2_SYNC:
2151         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2152                 config.rx_filter = HWTSTAMP_FILTER_ALL;
2153                 break;
2154         default:
2155                 return -ERANGE;
2156         }
2157
2158         if (mlx4_en_timestamp_config(dev, config.tx_type, config.rx_filter)) {
2159                 config.tx_type = HWTSTAMP_TX_OFF;
2160                 config.rx_filter = HWTSTAMP_FILTER_NONE;
2161         }
2162
2163         return copy_to_user(ifr->ifr_data, &config,
2164                             sizeof(config)) ? -EFAULT : 0;
2165 }
2166
2167 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2168 {
2169         struct mlx4_en_priv *priv = netdev_priv(dev);
2170
2171         return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2172                             sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2173 }
2174
2175 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2176 {
2177         switch (cmd) {
2178         case SIOCSHWTSTAMP:
2179                 return mlx4_en_hwtstamp_set(dev, ifr);
2180         case SIOCGHWTSTAMP:
2181                 return mlx4_en_hwtstamp_get(dev, ifr);
2182         default:
2183                 return -EOPNOTSUPP;
2184         }
2185 }
2186
2187 static int mlx4_en_set_features(struct net_device *netdev,
2188                 netdev_features_t features)
2189 {
2190         struct mlx4_en_priv *priv = netdev_priv(netdev);
2191
2192         if (features & NETIF_F_LOOPBACK)
2193                 priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
2194         else
2195                 priv->ctrl_flags &=
2196                         cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
2197
2198         mlx4_en_update_loopback_state(netdev, features);
2199
2200         return 0;
2201
2202 }
2203
2204 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2205 {
2206         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2207         struct mlx4_en_dev *mdev = en_priv->mdev;
2208         u64 mac_u64 = mlx4_en_mac_to_u64(mac);
2209
2210         if (!is_valid_ether_addr(mac))
2211                 return -EINVAL;
2212
2213         return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2214 }
2215
2216 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos)
2217 {
2218         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2219         struct mlx4_en_dev *mdev = en_priv->mdev;
2220
2221         return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos);
2222 }
2223
2224 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2225 {
2226         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2227         struct mlx4_en_dev *mdev = en_priv->mdev;
2228
2229         return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2230 }
2231
2232 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2233 {
2234         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2235         struct mlx4_en_dev *mdev = en_priv->mdev;
2236
2237         return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2238 }
2239
2240 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2241 {
2242         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2243         struct mlx4_en_dev *mdev = en_priv->mdev;
2244
2245         return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2246 }
2247
2248 #define PORT_ID_BYTE_LEN 8
2249 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2250                                     struct netdev_phys_port_id *ppid)
2251 {
2252         struct mlx4_en_priv *priv = netdev_priv(dev);
2253         struct mlx4_dev *mdev = priv->mdev->dev;
2254         int i;
2255         u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2256
2257         if (!phys_port_id)
2258                 return -EOPNOTSUPP;
2259
2260         ppid->id_len = sizeof(phys_port_id);
2261         for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2262                 ppid->id[i] =  phys_port_id & 0xff;
2263                 phys_port_id >>= 8;
2264         }
2265         return 0;
2266 }
2267
2268 static const struct net_device_ops mlx4_netdev_ops = {
2269         .ndo_open               = mlx4_en_open,
2270         .ndo_stop               = mlx4_en_close,
2271         .ndo_start_xmit         = mlx4_en_xmit,
2272         .ndo_select_queue       = mlx4_en_select_queue,
2273         .ndo_get_stats          = mlx4_en_get_stats,
2274         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2275         .ndo_set_mac_address    = mlx4_en_set_mac,
2276         .ndo_validate_addr      = eth_validate_addr,
2277         .ndo_change_mtu         = mlx4_en_change_mtu,
2278         .ndo_do_ioctl           = mlx4_en_ioctl,
2279         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2280         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2281         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2282 #ifdef CONFIG_NET_POLL_CONTROLLER
2283         .ndo_poll_controller    = mlx4_en_netpoll,
2284 #endif
2285         .ndo_set_features       = mlx4_en_set_features,
2286         .ndo_setup_tc           = mlx4_en_setup_tc,
2287 #ifdef CONFIG_RFS_ACCEL
2288         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2289 #endif
2290 #ifdef CONFIG_NET_RX_BUSY_POLL
2291         .ndo_busy_poll          = mlx4_en_low_latency_recv,
2292 #endif
2293         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2294 };
2295
2296 static const struct net_device_ops mlx4_netdev_ops_master = {
2297         .ndo_open               = mlx4_en_open,
2298         .ndo_stop               = mlx4_en_close,
2299         .ndo_start_xmit         = mlx4_en_xmit,
2300         .ndo_select_queue       = mlx4_en_select_queue,
2301         .ndo_get_stats          = mlx4_en_get_stats,
2302         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2303         .ndo_set_mac_address    = mlx4_en_set_mac,
2304         .ndo_validate_addr      = eth_validate_addr,
2305         .ndo_change_mtu         = mlx4_en_change_mtu,
2306         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2307         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2308         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2309         .ndo_set_vf_mac         = mlx4_en_set_vf_mac,
2310         .ndo_set_vf_vlan        = mlx4_en_set_vf_vlan,
2311         .ndo_set_vf_spoofchk    = mlx4_en_set_vf_spoofchk,
2312         .ndo_set_vf_link_state  = mlx4_en_set_vf_link_state,
2313         .ndo_get_vf_config      = mlx4_en_get_vf_config,
2314 #ifdef CONFIG_NET_POLL_CONTROLLER
2315         .ndo_poll_controller    = mlx4_en_netpoll,
2316 #endif
2317         .ndo_set_features       = mlx4_en_set_features,
2318         .ndo_setup_tc           = mlx4_en_setup_tc,
2319 #ifdef CONFIG_RFS_ACCEL
2320         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2321 #endif
2322         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2323 };
2324
2325 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2326                         struct mlx4_en_port_profile *prof)
2327 {
2328         struct net_device *dev;
2329         struct mlx4_en_priv *priv;
2330         int i;
2331         int err;
2332         u64 mac_u64;
2333
2334         dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2335                                  MAX_TX_RINGS, MAX_RX_RINGS);
2336         if (dev == NULL)
2337                 return -ENOMEM;
2338
2339         netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2340         netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2341
2342         SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
2343         dev->dev_id =  port - 1;
2344
2345         /*
2346          * Initialize driver private data
2347          */
2348
2349         priv = netdev_priv(dev);
2350         memset(priv, 0, sizeof(struct mlx4_en_priv));
2351         priv->dev = dev;
2352         priv->mdev = mdev;
2353         priv->ddev = &mdev->pdev->dev;
2354         priv->prof = prof;
2355         priv->port = port;
2356         priv->port_up = false;
2357         priv->flags = prof->flags;
2358         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2359                         MLX4_WQE_CTRL_SOLICITED);
2360         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2361         priv->tx_ring_num = prof->tx_ring_num;
2362
2363         priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring *) * MAX_TX_RINGS,
2364                                 GFP_KERNEL);
2365         if (!priv->tx_ring) {
2366                 err = -ENOMEM;
2367                 goto out;
2368         }
2369         priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq *) * MAX_TX_RINGS,
2370                               GFP_KERNEL);
2371         if (!priv->tx_cq) {
2372                 err = -ENOMEM;
2373                 goto out;
2374         }
2375         priv->rx_ring_num = prof->rx_ring_num;
2376         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2377         priv->mac_index = -1;
2378         priv->msg_enable = MLX4_EN_MSG_LEVEL;
2379         spin_lock_init(&priv->stats_lock);
2380         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2381         INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2382         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2383         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2384         INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2385 #ifdef CONFIG_MLX4_EN_DCB
2386         if (!mlx4_is_slave(priv->mdev->dev)) {
2387                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
2388                         dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2389                 } else {
2390                         en_info(priv, "enabling only PFC DCB ops\n");
2391                         dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2392                 }
2393         }
2394 #endif
2395
2396         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2397                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2398
2399         /* Query for default mac and max mtu */
2400         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2401
2402         /* Set default MAC */
2403         dev->addr_len = ETH_ALEN;
2404         mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2405         if (!is_valid_ether_addr(dev->dev_addr)) {
2406                 if (mlx4_is_slave(priv->mdev->dev)) {
2407                         eth_hw_addr_random(dev);
2408                         en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
2409                         mac_u64 = mlx4_en_mac_to_u64(dev->dev_addr);
2410                         mdev->dev->caps.def_mac[priv->port] = mac_u64;
2411                 } else {
2412                         en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2413                                priv->port, dev->dev_addr);
2414                         err = -EINVAL;
2415                         goto out;
2416                 }
2417         }
2418
2419         memcpy(priv->prev_mac, dev->dev_addr, sizeof(priv->prev_mac));
2420
2421         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2422                                           DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2423         err = mlx4_en_alloc_resources(priv);
2424         if (err)
2425                 goto out;
2426
2427 #ifdef CONFIG_RFS_ACCEL
2428         INIT_LIST_HEAD(&priv->filters);
2429         spin_lock_init(&priv->filters_lock);
2430 #endif
2431
2432         /* Initialize time stamping config */
2433         priv->hwtstamp_config.flags = 0;
2434         priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
2435         priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2436
2437         /* Allocate page for receive rings */
2438         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2439                                 MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2440         if (err) {
2441                 en_err(priv, "Failed to allocate page for rx qps\n");
2442                 goto out;
2443         }
2444         priv->allocated = 1;
2445
2446         /*
2447          * Initialize netdev entry points
2448          */
2449         if (mlx4_is_master(priv->mdev->dev))
2450                 dev->netdev_ops = &mlx4_netdev_ops_master;
2451         else
2452                 dev->netdev_ops = &mlx4_netdev_ops;
2453         dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2454         netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2455         netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2456
2457         SET_ETHTOOL_OPS(dev, &mlx4_en_ethtool_ops);
2458
2459         /*
2460          * Set driver features
2461          */
2462         dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2463         if (mdev->LSO_support)
2464                 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2465
2466         dev->vlan_features = dev->hw_features;
2467
2468         dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2469         dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2470                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
2471                         NETIF_F_HW_VLAN_CTAG_FILTER;
2472         dev->hw_features |= NETIF_F_LOOPBACK;
2473
2474         if (mdev->dev->caps.steering_mode ==
2475             MLX4_STEERING_MODE_DEVICE_MANAGED)
2476                 dev->hw_features |= NETIF_F_NTUPLE;
2477
2478         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
2479                 dev->priv_flags |= IFF_UNICAST_FLT;
2480
2481         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2482                 dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
2483                                         NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL;
2484                 dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
2485                 dev->features    |= NETIF_F_GSO_UDP_TUNNEL;
2486         }
2487
2488         mdev->pndev[port] = dev;
2489
2490         netif_carrier_off(dev);
2491         mlx4_en_set_default_moderation(priv);
2492
2493         err = register_netdev(dev);
2494         if (err) {
2495                 en_err(priv, "Netdev registration failed for port %d\n", port);
2496                 goto out;
2497         }
2498         priv->registered = 1;
2499
2500         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2501         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2502
2503         mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
2504
2505         /* Configure port */
2506         mlx4_en_calc_rx_buf(dev);
2507         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2508                                     priv->rx_skb_size + ETH_FCS_LEN,
2509                                     prof->tx_pause, prof->tx_ppp,
2510                                     prof->rx_pause, prof->rx_ppp);
2511         if (err) {
2512                 en_err(priv, "Failed setting port general configurations "
2513                        "for port %d, with error %d\n", priv->port, err);
2514                 goto out;
2515         }
2516
2517         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
2518                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC);
2519                 if (err) {
2520                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
2521                                err);
2522                         goto out;
2523                 }
2524         }
2525
2526         /* Init port */
2527         en_warn(priv, "Initializing port\n");
2528         err = mlx4_INIT_PORT(mdev->dev, priv->port);
2529         if (err) {
2530                 en_err(priv, "Failed Initializing port\n");
2531                 goto out;
2532         }
2533         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2534
2535         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2536                 queue_delayed_work(mdev->workqueue, &priv->service_task,
2537                                    SERVICE_TASK_DELAY);
2538
2539         return 0;
2540
2541 out:
2542         mlx4_en_destroy_netdev(dev);
2543         return err;
2544 }
2545