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
  42 #include <linux/mlx4/driver.h>
  43 #include <linux/mlx4/device.h>
  44 #include <linux/mlx4/cmd.h>
  45 #include <linux/mlx4/cq.h>
  46
  47 #include "mlx4_en.h"
  48 #include "en_port.h"
  49
  50 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
  51 {
  52         struct mlx4_en_priv *priv = netdev_priv(dev);
  53         int i;
  54         unsigned int offset = 0;
  55
  56         if (up && up != MLX4_EN_NUM_UP)
  57                 return -EINVAL;
  58
  59         netdev_set_num_tc(dev, up);
  60
  61         /* Partition Tx queues evenly amongst UP's */
  62         for (i = 0; i < up; i++) {
  63                 netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
  64                 offset += priv->num_tx_rings_p_up;
  65         }
  66
  67         return 0;
  68 }
  69
  70 #ifdef CONFIG_RFS_ACCEL
  71
  72 struct mlx4_en_filter {
  73         struct list_head next;
  74         struct work_struct work;
  75
  76         __be32 src_ip;
  77         __be32 dst_ip;
  78         __be16 src_port;
  79         __be16 dst_port;
  80
  81         int rxq_index;
  82         struct mlx4_en_priv *priv;
  83         u32 flow_id;                    /* RFS infrastructure id */
  84         int id;                         /* mlx4_en driver id */
  85         u64 reg_id;                     /* Flow steering API id */
  86         u8 activated;                   /* Used to prevent expiry before filter
  87                                          * is attached
  88                                          */
  89         struct hlist_node filter_chain;
  90 };
  91
  92 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
  93
  94 static void mlx4_en_filter_work(struct work_struct *work)
  95 {
  96         struct mlx4_en_filter *filter = container_of(work,
  97                                                      struct mlx4_en_filter,
  98                                                      work);
  99         struct mlx4_en_priv *priv = filter->priv;
 100         struct mlx4_spec_list spec_tcp = {
 101                 .id = MLX4_NET_TRANS_RULE_ID_TCP,
 102                 {
 103                         .tcp_udp = {
 104                                 .dst_port = filter->dst_port,
 105                                 .dst_port_msk = (__force __be16)-1,
 106                                 .src_port = filter->src_port,
 107                                 .src_port_msk = (__force __be16)-1,
 108                         },
 109                 },
 110         };
 111         struct mlx4_spec_list spec_ip = {
 112                 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
 113                 {
 114                         .ipv4 = {
 115                                 .dst_ip = filter->dst_ip,
 116                                 .dst_ip_msk = (__force __be32)-1,
 117                                 .src_ip = filter->src_ip,
 118                                 .src_ip_msk = (__force __be32)-1,
 119                         },
 120                 },
 121         };
 122         struct mlx4_spec_list spec_eth = {
 123                 .id = MLX4_NET_TRANS_RULE_ID_ETH,
 124         };
 125         struct mlx4_net_trans_rule rule = {
 126                 .list = LIST_HEAD_INIT(rule.list),
 127                 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
 128                 .exclusive = 1,
 129                 .allow_loopback = 1,
 130                 .promisc_mode = MLX4_FS_PROMISC_NONE,
 131                 .port = priv->port,
 132                 .priority = MLX4_DOMAIN_RFS,
 133         };
 134         int rc;
 135         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 136
 137         list_add_tail(&spec_eth.list, &rule.list);
 138         list_add_tail(&spec_ip.list, &rule.list);
 139         list_add_tail(&spec_tcp.list, &rule.list);
 140
 141         rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
 142         memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
 143         memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 144
 145         filter->activated = 0;
 146
 147         if (filter->reg_id) {
 148                 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 149                 if (rc && rc != -ENOENT)
 150                         en_err(priv, "Error detaching flow. rc = %d\n", rc);
 151         }
 152
 153         rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
 154         if (rc)
 155                 en_err(priv, "Error attaching flow. err = %d\n", rc);
 156
 157         mlx4_en_filter_rfs_expire(priv);
 158
 159         filter->activated = 1;
 160 }
 161
 162 static inline struct hlist_head *
 163 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 164                    __be16 src_port, __be16 dst_port)
 165 {
 166         unsigned long l;
 167         int bucket_idx;
 168
 169         l = (__force unsigned long)src_port |
 170             ((__force unsigned long)dst_port << 2);
 171         l ^= (__force unsigned long)(src_ip ^ dst_ip);
 172
 173         bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
 174
 175         return &priv->filter_hash[bucket_idx];
 176 }
 177
 178 static struct mlx4_en_filter *
 179 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
 180                      __be32 dst_ip, __be16 src_port, __be16 dst_port,
 181                      u32 flow_id)
 182 {
 183         struct mlx4_en_filter *filter = NULL;
 184
 185         filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
 186         if (!filter)
 187                 return NULL;
 188
 189         filter->priv = priv;
 190         filter->rxq_index = rxq_index;
 191         INIT_WORK(&filter->work, mlx4_en_filter_work);
 192
 193         filter->src_ip = src_ip;
 194         filter->dst_ip = dst_ip;
 195         filter->src_port = src_port;
 196         filter->dst_port = dst_port;
 197
 198         filter->flow_id = flow_id;
 199
 200         filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
 201
 202         list_add_tail(&filter->next, &priv->filters);
 203         hlist_add_head(&filter->filter_chain,
 204                        filter_hash_bucket(priv, src_ip, dst_ip, src_port,
 205                                           dst_port));
 206
 207         return filter;
 208 }
 209
 210 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
 211 {
 212         struct mlx4_en_priv *priv = filter->priv;
 213         int rc;
 214
 215         list_del(&filter->next);
 216
 217         rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 218         if (rc && rc != -ENOENT)
 219                 en_err(priv, "Error detaching flow. rc = %d\n", rc);
 220
 221         kfree(filter);
 222 }
 223
 224 static inline struct mlx4_en_filter *
 225 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 226                     __be16 src_port, __be16 dst_port)
 227 {
 228         struct hlist_node *elem;
 229         struct mlx4_en_filter *filter;
 230         struct mlx4_en_filter *ret = NULL;
 231
 232         hlist_for_each_entry(filter, elem,
 233                              filter_hash_bucket(priv, src_ip, dst_ip,
 234                                                 src_port, dst_port),
 235                              filter_chain) {
 236                 if (filter->src_ip == src_ip &&
 237                     filter->dst_ip == dst_ip &&
 238                     filter->src_port == src_port &&
 239                     filter->dst_port == dst_port) {
 240                         ret = filter;
 241                         break;
 242                 }
 243         }
 244
 245         return ret;
 246 }
 247
 248 static int
 249 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
 250                    u16 rxq_index, u32 flow_id)
 251 {
 252         struct mlx4_en_priv *priv = netdev_priv(net_dev);
 253         struct mlx4_en_filter *filter;
 254         const struct iphdr *ip;
 255         const __be16 *ports;
 256         __be32 src_ip;
 257         __be32 dst_ip;
 258         __be16 src_port;
 259         __be16 dst_port;
 260         int nhoff = skb_network_offset(skb);
 261         int ret = 0;
 262
 263         if (skb->protocol != htons(ETH_P_IP))
 264                 return -EPROTONOSUPPORT;
 265
 266         ip = (const struct iphdr *)(skb->data + nhoff);
 267         if (ip_is_fragment(ip))
 268                 return -EPROTONOSUPPORT;
 269
 270         ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
 271
 272         src_ip = ip->saddr;
 273         dst_ip = ip->daddr;
 274         src_port = ports[0];
 275         dst_port = ports[1];
 276
 277         if (ip->protocol != IPPROTO_TCP)
 278                 return -EPROTONOSUPPORT;
 279
 280         spin_lock_bh(&priv->filters_lock);
 281         filter = mlx4_en_filter_find(priv, src_ip, dst_ip, src_port, dst_port);
 282         if (filter) {
 283                 if (filter->rxq_index == rxq_index)
 284                         goto out;
 285
 286                 filter->rxq_index = rxq_index;
 287         } else {
 288                 filter = mlx4_en_filter_alloc(priv, rxq_index,
 289                                               src_ip, dst_ip,
 290                                               src_port, dst_port, flow_id);
 291                 if (!filter) {
 292                         ret = -ENOMEM;
 293                         goto err;
 294                 }
 295         }
 296
 297         queue_work(priv->mdev->workqueue, &filter->work);
 298
 299 out:
 300         ret = filter->id;
 301 err:
 302         spin_unlock_bh(&priv->filters_lock);
 303
 304         return ret;
 305 }
 306
 307 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv,
 308                              struct mlx4_en_rx_ring *rx_ring)
 309 {
 310         struct mlx4_en_filter *filter, *tmp;
 311         LIST_HEAD(del_list);
 312
 313         spin_lock_bh(&priv->filters_lock);
 314         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 315                 list_move(&filter->next, &del_list);
 316                 hlist_del(&filter->filter_chain);
 317         }
 318         spin_unlock_bh(&priv->filters_lock);
 319
 320         list_for_each_entry_safe(filter, tmp, &del_list, next) {
 321                 cancel_work_sync(&filter->work);
 322                 mlx4_en_filter_free(filter);
 323         }
 324 }
 325
 326 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
 327 {
 328         struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
 329         LIST_HEAD(del_list);
 330         int i = 0;
 331
 332         spin_lock_bh(&priv->filters_lock);
 333         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 334                 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
 335                         break;
 336
 337                 if (filter->activated &&
 338                     !work_pending(&filter->work) &&
 339                     rps_may_expire_flow(priv->dev,
 340                                         filter->rxq_index, filter->flow_id,
 341                                         filter->id)) {
 342                         list_move(&filter->next, &del_list);
 343                         hlist_del(&filter->filter_chain);
 344                 } else
 345                         last_filter = filter;
 346
 347                 i++;
 348         }
 349
 350         if (last_filter && (&last_filter->next != priv->filters.next))
 351                 list_move(&priv->filters, &last_filter->next);
 352
 353         spin_unlock_bh(&priv->filters_lock);
 354
 355         list_for_each_entry_safe(filter, tmp, &del_list, next)
 356                 mlx4_en_filter_free(filter);
 357 }
 358 #endif
 359
 360 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev, unsigned short vid)
 361 {
 362         struct mlx4_en_priv *priv = netdev_priv(dev);
 363         struct mlx4_en_dev *mdev = priv->mdev;
 364         int err;
 365         int idx;
 366
 367         en_dbg(HW, priv, "adding VLAN:%d\n", vid);
 368
 369         set_bit(vid, priv->active_vlans);
 370
 371         /* Add VID to port VLAN filter */
 372         mutex_lock(&mdev->state_lock);
 373         if (mdev->device_up && priv->port_up) {
 374                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 375                 if (err)
 376                         en_err(priv, "Failed configuring VLAN filter\n");
 377         }
 378         if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
 379                 en_err(priv, "failed adding vlan %d\n", vid);
 380         mutex_unlock(&mdev->state_lock);
 381
 382         return 0;
 383 }
 384
 385 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
 386 {
 387         struct mlx4_en_priv *priv = netdev_priv(dev);
 388         struct mlx4_en_dev *mdev = priv->mdev;
 389         int err;
 390         int idx;
 391
 392         en_dbg(HW, priv, "Killing VID:%d\n", vid);
 393
 394         clear_bit(vid, priv->active_vlans);
 395
 396         /* Remove VID from port VLAN filter */
 397         mutex_lock(&mdev->state_lock);
 398         if (!mlx4_find_cached_vlan(mdev->dev, priv->port, vid, &idx))
 399                 mlx4_unregister_vlan(mdev->dev, priv->port, idx);
 400         else
 401                 en_err(priv, "could not find vid %d in cache\n", vid);
 402
 403         if (mdev->device_up && priv->port_up) {
 404                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 405                 if (err)
 406                         en_err(priv, "Failed configuring VLAN filter\n");
 407         }
 408         mutex_unlock(&mdev->state_lock);
 409
 410         return 0;
 411 }
 412
 413 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
 414 {
 415         unsigned int i;
 416         for (i = ETH_ALEN - 1; i; --i) {
 417                 dst_mac[i] = src_mac & 0xff;
 418                 src_mac >>= 8;
 419         }
 420         memset(&dst_mac[ETH_ALEN], 0, 2);
 421 }
 422
 423 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
 424                                 unsigned char *mac, int *qpn, u64 *reg_id)
 425 {
 426         struct mlx4_en_dev *mdev = priv->mdev;
 427         struct mlx4_dev *dev = mdev->dev;
 428         int err;
 429
 430         switch (dev->caps.steering_mode) {
 431         case MLX4_STEERING_MODE_B0: {
 432                 struct mlx4_qp qp;
 433                 u8 gid[16] = {0};
 434
 435                 qp.qpn = *qpn;
 436                 memcpy(&gid[10], mac, ETH_ALEN);
 437                 gid[5] = priv->port;
 438
 439                 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
 440                 break;
 441         }
 442         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 443                 struct mlx4_spec_list spec_eth = { {NULL} };
 444                 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 445
 446                 struct mlx4_net_trans_rule rule = {
 447                         .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 448                         .exclusive = 0,
 449                         .allow_loopback = 1,
 450                         .promisc_mode = MLX4_FS_PROMISC_NONE,
 451                         .priority = MLX4_DOMAIN_NIC,
 452                 };
 453
 454                 rule.port = priv->port;
 455                 rule.qpn = *qpn;
 456                 INIT_LIST_HEAD(&rule.list);
 457
 458                 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
 459                 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
 460                 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 461                 list_add_tail(&spec_eth.list, &rule.list);
 462
 463                 err = mlx4_flow_attach(dev, &rule, reg_id);
 464                 break;
 465         }
 466         default:
 467                 return -EINVAL;
 468         }
 469         if (err)
 470                 en_warn(priv, "Failed Attaching Unicast\n");
 471
 472         return err;
 473 }
 474
 475 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
 476                                      unsigned char *mac, int qpn, u64 reg_id)
 477 {
 478         struct mlx4_en_dev *mdev = priv->mdev;
 479         struct mlx4_dev *dev = mdev->dev;
 480
 481         switch (dev->caps.steering_mode) {
 482         case MLX4_STEERING_MODE_B0: {
 483                 struct mlx4_qp qp;
 484                 u8 gid[16] = {0};
 485
 486                 qp.qpn = qpn;
 487                 memcpy(&gid[10], mac, ETH_ALEN);
 488                 gid[5] = priv->port;
 489
 490                 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
 491                 break;
 492         }
 493         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 494                 mlx4_flow_detach(dev, reg_id);
 495                 break;
 496         }
 497         default:
 498                 en_err(priv, "Invalid steering mode.\n");
 499         }
 500 }
 501
 502 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
 503 {
 504         struct mlx4_en_dev *mdev = priv->mdev;
 505         struct mlx4_dev *dev = mdev->dev;
 506         struct mlx4_mac_entry *entry;
 507         int index = 0;
 508         int err = 0;
 509         u64 reg_id;
 510         int *qpn = &priv->base_qpn;
 511         u64 mac = mlx4_en_mac_to_u64(priv->dev->dev_addr);
 512
 513         en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
 514                priv->dev->dev_addr);
 515         index = mlx4_register_mac(dev, priv->port, mac);
 516         if (index < 0) {
 517                 err = index;
 518                 en_err(priv, "Failed adding MAC: %pM\n",
 519                        priv->dev->dev_addr);
 520                 return err;
 521         }
 522
 523         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 524                 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
 525                 *qpn = base_qpn + index;
 526                 return 0;
 527         }
 528
 529         err = mlx4_qp_reserve_range(dev, 1, 1, qpn);
 530         en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
 531         if (err) {
 532                 en_err(priv, "Failed to reserve qp for mac registration\n");
 533                 goto qp_err;
 534         }
 535
 536         err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
 537         if (err)
 538                 goto steer_err;
 539
 540         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 541         if (!entry) {
 542                 err = -ENOMEM;
 543                 goto alloc_err;
 544         }
 545         memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
 546         entry->reg_id = reg_id;
 547
 548         hlist_add_head_rcu(&entry->hlist,
 549                            &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
 550
 551         return 0;
 552
 553 alloc_err:
 554         mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
 555
 556 steer_err:
 557         mlx4_qp_release_range(dev, *qpn, 1);
 558
 559 qp_err:
 560         mlx4_unregister_mac(dev, priv->port, mac);
 561         return err;
 562 }
 563
 564 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
 565 {
 566         struct mlx4_en_dev *mdev = priv->mdev;
 567         struct mlx4_dev *dev = mdev->dev;
 568         int qpn = priv->base_qpn;
 569         u64 mac = mlx4_en_mac_to_u64(priv->dev->dev_addr);
 570
 571         en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 572                priv->dev->dev_addr);
 573         mlx4_unregister_mac(dev, priv->port, mac);
 574
 575         if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
 576                 struct mlx4_mac_entry *entry;
 577                 struct hlist_node *n, *tmp;
 578                 struct hlist_head *bucket;
 579                 unsigned int mac_hash;
 580
 581                 mac_hash = priv->dev->dev_addr[MLX4_EN_MAC_HASH_IDX];
 582                 bucket = &priv->mac_hash[mac_hash];
 583                 hlist_for_each_entry_safe(entry, n, tmp, bucket, hlist) {
 584                         if (ether_addr_equal_64bits(entry->mac,
 585                                                     priv->dev->dev_addr)) {
 586                                 en_dbg(DRV, priv, "Releasing qp: port %d, MAC %pM, qpn %d\n",
 587                                        priv->port, priv->dev->dev_addr, qpn);
 588                                 mlx4_en_uc_steer_release(priv, entry->mac,
 589                                                          qpn, entry->reg_id);
 590                                 mlx4_qp_release_range(dev, qpn, 1);
 591
 592                                 hlist_del_rcu(&entry->hlist);
 593                                 kfree_rcu(entry, rcu);
 594                                 break;
 595                         }
 596                 }
 597         }
 598 }
 599
 600 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
 601                                unsigned char *new_mac, unsigned char *prev_mac)
 602 {
 603         struct mlx4_en_dev *mdev = priv->mdev;
 604         struct mlx4_dev *dev = mdev->dev;
 605         int err = 0;
 606         u64 new_mac_u64 = mlx4_en_mac_to_u64(new_mac);
 607
 608         if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
 609                 struct hlist_head *bucket;
 610                 unsigned int mac_hash;
 611                 struct mlx4_mac_entry *entry;
 612                 struct hlist_node *n, *tmp;
 613                 u64 prev_mac_u64 = mlx4_en_mac_to_u64(prev_mac);
 614
 615                 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
 616                 hlist_for_each_entry_safe(entry, n, tmp, bucket, hlist) {
 617                         if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
 618                                 mlx4_en_uc_steer_release(priv, entry->mac,
 619                                                          qpn, entry->reg_id);
 620                                 mlx4_unregister_mac(dev, priv->port,
 621                                                     prev_mac_u64);
 622                                 hlist_del_rcu(&entry->hlist);
 623                                 synchronize_rcu();
 624                                 memcpy(entry->mac, new_mac, ETH_ALEN);
 625                                 entry->reg_id = 0;
 626                                 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
 627                                 hlist_add_head_rcu(&entry->hlist,
 628                                                    &priv->mac_hash[mac_hash]);
 629                                 mlx4_register_mac(dev, priv->port, new_mac_u64);
 630                                 err = mlx4_en_uc_steer_add(priv, new_mac,
 631                                                            &qpn,
 632                                                            &entry->reg_id);
 633                                 return err;
 634                         }
 635                 }
 636                 return -EINVAL;
 637         }
 638
 639         return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
 640 }
 641
 642 u64 mlx4_en_mac_to_u64(u8 *addr)
 643 {
 644         u64 mac = 0;
 645         int i;
 646
 647         for (i = 0; i < ETH_ALEN; i++) {
 648                 mac <<= 8;
 649                 mac |= addr[i];
 650         }
 651         return mac;
 652 }
 653
 654 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
 655 {
 656         struct mlx4_en_priv *priv = netdev_priv(dev);
 657         struct mlx4_en_dev *mdev = priv->mdev;
 658         struct sockaddr *saddr = addr;
 659
 660         if (!is_valid_ether_addr(saddr->sa_data))
 661                 return -EADDRNOTAVAIL;
 662
 663         memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
 664         queue_work(mdev->workqueue, &priv->mac_task);
 665         return 0;
 666 }
 667
 668 static void mlx4_en_do_set_mac(struct work_struct *work)
 669 {
 670         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
 671                                                  mac_task);
 672         struct mlx4_en_dev *mdev = priv->mdev;
 673         int err = 0;
 674
 675         mutex_lock(&mdev->state_lock);
 676         if (priv->port_up) {
 677                 /* Remove old MAC and insert the new one */
 678                 err = mlx4_en_replace_mac(priv, priv->base_qpn,
 679                                           priv->dev->dev_addr, priv->prev_mac);
 680                 if (err)
 681                         en_err(priv, "Failed changing HW MAC address\n");
 682                 memcpy(priv->prev_mac, priv->dev->dev_addr,
 683                        sizeof(priv->prev_mac));
 684         } else
 685                 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
 686
 687         mutex_unlock(&mdev->state_lock);
 688 }
 689
 690 static void mlx4_en_clear_list(struct net_device *dev)
 691 {
 692         struct mlx4_en_priv *priv = netdev_priv(dev);
 693         struct mlx4_en_mc_list *tmp, *mc_to_del;
 694
 695         list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
 696                 list_del(&mc_to_del->list);
 697                 kfree(mc_to_del);
 698         }
 699 }
 700
 701 static void mlx4_en_cache_mclist(struct net_device *dev)
 702 {
 703         struct mlx4_en_priv *priv = netdev_priv(dev);
 704         struct netdev_hw_addr *ha;
 705         struct mlx4_en_mc_list *tmp;
 706
 707         mlx4_en_clear_list(dev);
 708         netdev_for_each_mc_addr(ha, dev) {
 709                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 710                 if (!tmp) {
 711                         mlx4_en_clear_list(dev);
 712                         return;
 713                 }
 714                 memcpy(tmp->addr, ha->addr, ETH_ALEN);
 715                 list_add_tail(&tmp->list, &priv->mc_list);
 716         }
 717 }
 718
 719 static void update_mclist_flags(struct mlx4_en_priv *priv,
 720                                 struct list_head *dst,
 721                                 struct list_head *src)
 722 {
 723         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 724         bool found;
 725
 726         /* Find all the entries that should be removed from dst,
 727          * These are the entries that are not found in src
 728          */
 729         list_for_each_entry(dst_tmp, dst, list) {
 730                 found = false;
 731                 list_for_each_entry(src_tmp, src, list) {
 732                         if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
 733                                 found = true;
 734                                 break;
 735                         }
 736                 }
 737                 if (!found)
 738                         dst_tmp->action = MCLIST_REM;
 739         }
 740
 741         /* Add entries that exist in src but not in dst
 742          * mark them as need to add
 743          */
 744         list_for_each_entry(src_tmp, src, list) {
 745                 found = false;
 746                 list_for_each_entry(dst_tmp, dst, list) {
 747                         if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
 748                                 dst_tmp->action = MCLIST_NONE;
 749                                 found = true;
 750                                 break;
 751                         }
 752                 }
 753                 if (!found) {
 754                         new_mc = kmemdup(src_tmp,
 755                                          sizeof(struct mlx4_en_mc_list),
 756                                          GFP_KERNEL);
 757                         if (!new_mc)
 758                                 return;
 759
 760                         new_mc->action = MCLIST_ADD;
 761                         list_add_tail(&new_mc->list, dst);
 762                 }
 763         }
 764 }
 765
 766 static void mlx4_en_set_rx_mode(struct net_device *dev)
 767 {
 768         struct mlx4_en_priv *priv = netdev_priv(dev);
 769
 770         if (!priv->port_up)
 771                 return;
 772
 773         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 774 }
 775
 776 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 777                                      struct mlx4_en_dev *mdev)
 778 {
 779         int err = 0;
 780
 781         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 782                 if (netif_msg_rx_status(priv))
 783                         en_warn(priv, "Entering promiscuous mode\n");
 784                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 785
 786                 /* Enable promiscouos mode */
 787                 switch (mdev->dev->caps.steering_mode) {
 788                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 789                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 790                                                           priv->port,
 791                                                           priv->base_qpn,
 792                                                           MLX4_FS_PROMISC_UPLINK);
 793                         if (err)
 794                                 en_err(priv, "Failed enabling promiscuous mode\n");
 795                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 796                         break;
 797
 798                 case MLX4_STEERING_MODE_B0:
 799                         err = mlx4_unicast_promisc_add(mdev->dev,
 800                                                        priv->base_qpn,
 801                                                        priv->port);
 802                         if (err)
 803                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 804
 805                         /* Add the default qp number as multicast
 806                          * promisc
 807                          */
 808                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 809                                 err = mlx4_multicast_promisc_add(mdev->dev,
 810                                                                  priv->base_qpn,
 811                                                                  priv->port);
 812                                 if (err)
 813                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 814                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 815                         }
 816                         break;
 817
 818                 case MLX4_STEERING_MODE_A0:
 819                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 820                                                      priv->port,
 821                                                      priv->base_qpn,
 822                                                      1);
 823                         if (err)
 824                                 en_err(priv, "Failed enabling promiscuous mode\n");
 825                         break;
 826                 }
 827
 828                 /* Disable port multicast filter (unconditionally) */
 829                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 830                                           0, MLX4_MCAST_DISABLE);
 831                 if (err)
 832                         en_err(priv, "Failed disabling multicast filter\n");
 833
 834                 /* Disable port VLAN filter */
 835                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 836                 if (err)
 837                         en_err(priv, "Failed disabling VLAN filter\n");
 838         }
 839 }
 840
 841 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 842                                        struct mlx4_en_dev *mdev)
 843 {
 844         int err = 0;
 845
 846         if (netif_msg_rx_status(priv))
 847                 en_warn(priv, "Leaving promiscuous mode\n");
 848         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 849
 850         /* Disable promiscouos mode */
 851         switch (mdev->dev->caps.steering_mode) {
 852         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 853                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 854                                                      priv->port,
 855                                                      MLX4_FS_PROMISC_UPLINK);
 856                 if (err)
 857                         en_err(priv, "Failed disabling promiscuous mode\n");
 858                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 859                 break;
 860
 861         case MLX4_STEERING_MODE_B0:
 862                 err = mlx4_unicast_promisc_remove(mdev->dev,
 863                                                   priv->base_qpn,
 864                                                   priv->port);
 865                 if (err)
 866                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 867                 /* Disable Multicast promisc */
 868                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 869                         err = mlx4_multicast_promisc_remove(mdev->dev,
 870                                                             priv->base_qpn,
 871                                                             priv->port);
 872                         if (err)
 873                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 874                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 875                 }
 876                 break;
 877
 878         case MLX4_STEERING_MODE_A0:
 879                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 880                                              priv->port,
 881                                              priv->base_qpn, 0);
 882                 if (err)
 883                         en_err(priv, "Failed disabling promiscuous mode\n");
 884                 break;
 885         }
 886
 887         /* Enable port VLAN filter */
 888         err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 889         if (err)
 890                 en_err(priv, "Failed enabling VLAN filter\n");
 891 }
 892
 893 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
 894                                  struct net_device *dev,
 895                                  struct mlx4_en_dev *mdev)
 896 {
 897         struct mlx4_en_mc_list *mclist, *tmp;
 898         u64 mcast_addr = 0;
 899         u8 mc_list[16] = {0};
 900         int err = 0;
 901
 902         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
 903         if (dev->flags & IFF_ALLMULTI) {
 904                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 905                                           0, MLX4_MCAST_DISABLE);
 906                 if (err)
 907                         en_err(priv, "Failed disabling multicast filter\n");
 908
 909                 /* Add the default qp number as multicast promisc */
 910                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 911                         switch (mdev->dev->caps.steering_mode) {
 912                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 913                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
 914                                                                   priv->port,
 915                                                                   priv->base_qpn,
 916                                                                   MLX4_FS_PROMISC_ALL_MULTI);
 917                                 break;
 918
 919                         case MLX4_STEERING_MODE_B0:
 920                                 err = mlx4_multicast_promisc_add(mdev->dev,
 921                                                                  priv->base_qpn,
 922                                                                  priv->port);
 923                                 break;
 924
 925                         case MLX4_STEERING_MODE_A0:
 926                                 break;
 927                         }
 928                         if (err)
 929                                 en_err(priv, "Failed entering multicast promisc mode\n");
 930                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 931                 }
 932         } else {
 933                 /* Disable Multicast promisc */
 934                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 935                         switch (mdev->dev->caps.steering_mode) {
 936                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 937                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 938                                                                      priv->port,
 939                                                                      MLX4_FS_PROMISC_ALL_MULTI);
 940                                 break;
 941
 942                         case MLX4_STEERING_MODE_B0:
 943                                 err = mlx4_multicast_promisc_remove(mdev->dev,
 944                                                                     priv->base_qpn,
 945                                                                     priv->port);
 946                                 break;
 947
 948                         case MLX4_STEERING_MODE_A0:
 949                                 break;
 950                         }
 951                         if (err)
 952                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 953                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 954                 }
 955
 956                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 957                                           0, MLX4_MCAST_DISABLE);
 958                 if (err)
 959                         en_err(priv, "Failed disabling multicast filter\n");
 960
 961                 /* Flush mcast filter and init it with broadcast address */
 962                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
 963                                     1, MLX4_MCAST_CONFIG);
 964
 965                 /* Update multicast list - we cache all addresses so they won't
 966                  * change while HW is updated holding the command semaphor */
 967                 netif_addr_lock_bh(dev);
 968                 mlx4_en_cache_mclist(dev);
 969                 netif_addr_unlock_bh(dev);
 970                 list_for_each_entry(mclist, &priv->mc_list, list) {
 971                         mcast_addr = mlx4_en_mac_to_u64(mclist->addr);
 972                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
 973                                             mcast_addr, 0, MLX4_MCAST_CONFIG);
 974                 }
 975                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 976                                           0, MLX4_MCAST_ENABLE);
 977                 if (err)
 978                         en_err(priv, "Failed enabling multicast filter\n");
 979
 980                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
 981                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
 982                         if (mclist->action == MCLIST_REM) {
 983                                 /* detach this address and delete from list */
 984                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
 985                                 mc_list[5] = priv->port;
 986                                 err = mlx4_multicast_detach(mdev->dev,
 987                                                             &priv->rss_map.indir_qp,
 988                                                             mc_list,
 989                                                             MLX4_PROT_ETH,
 990                                                             mclist->reg_id);
 991                                 if (err)
 992                                         en_err(priv, "Fail to detach multicast address\n");
 993
 994                                 /* remove from list */
 995                                 list_del(&mclist->list);
 996                                 kfree(mclist);
 997                         } else if (mclist->action == MCLIST_ADD) {
 998                                 /* attach the address */
 999                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1000                                 /* needed for B0 steering support */
1001                                 mc_list[5] = priv->port;
1002                                 err = mlx4_multicast_attach(mdev->dev,
1003                                                             &priv->rss_map.indir_qp,
1004                                                             mc_list,
1005                                                             priv->port, 0,
1006                                                             MLX4_PROT_ETH,
1007                                                             &mclist->reg_id);
1008                                 if (err)
1009                                         en_err(priv, "Fail to attach multicast address\n");
1010
1011                         }
1012                 }
1013         }
1014 }
1015
1016 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1017                                  struct net_device *dev,
1018                                  struct mlx4_en_dev *mdev)
1019 {
1020         struct netdev_hw_addr *ha;
1021         struct mlx4_mac_entry *entry;
1022         struct hlist_node *n, *tmp;
1023         bool found;
1024         u64 mac;
1025         int err = 0;
1026         struct hlist_head *bucket;
1027         unsigned int i;
1028         int removed = 0;
1029         u32 prev_flags;
1030
1031         /* Note that we do not need to protect our mac_hash traversal with rcu,
1032          * since all modification code is protected by mdev->state_lock
1033          */
1034
1035         /* find what to remove */
1036         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1037                 bucket = &priv->mac_hash[i];
1038                 hlist_for_each_entry_safe(entry, n, tmp, bucket, hlist) {
1039                         found = false;
1040                         netdev_for_each_uc_addr(ha, dev) {
1041                                 if (ether_addr_equal_64bits(entry->mac,
1042                                                             ha->addr)) {
1043                                         found = true;
1044                                         break;
1045                                 }
1046                         }
1047
1048                         /* MAC address of the port is not in uc list */
1049                         if (ether_addr_equal_64bits(entry->mac, dev->dev_addr))
1050                                 found = true;
1051
1052                         if (!found) {
1053                                 mac = mlx4_en_mac_to_u64(entry->mac);
1054                                 mlx4_en_uc_steer_release(priv, entry->mac,
1055                                                          priv->base_qpn,
1056                                                          entry->reg_id);
1057                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1058
1059                                 hlist_del_rcu(&entry->hlist);
1060                                 kfree_rcu(entry, rcu);
1061                                 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1062                                        entry->mac, priv->port);
1063                                 ++removed;
1064                         }
1065                 }
1066         }
1067
1068         /* if we didn't remove anything, there is no use in trying to add
1069          * again once we are in a forced promisc mode state
1070          */
1071         if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1072                 return;
1073
1074         prev_flags = priv->flags;
1075         priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1076
1077         /* find what to add */
1078         netdev_for_each_uc_addr(ha, dev) {
1079                 found = false;
1080                 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1081                 hlist_for_each_entry(entry, n, bucket, hlist) {
1082                         if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1083                                 found = true;
1084                                 break;
1085                         }
1086                 }
1087
1088                 if (!found) {
1089                         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1090                         if (!entry) {
1091                                 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1092                                        ha->addr, priv->port);
1093                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1094                                 break;
1095                         }
1096                         mac = mlx4_en_mac_to_u64(ha->addr);
1097                         memcpy(entry->mac, ha->addr, ETH_ALEN);
1098                         err = mlx4_register_mac(mdev->dev, priv->port, mac);
1099                         if (err < 0) {
1100                                 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1101                                        ha->addr, priv->port, err);
1102                                 kfree(entry);
1103                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1104                                 break;
1105                         }
1106                         err = mlx4_en_uc_steer_add(priv, ha->addr,
1107                                                    &priv->base_qpn,
1108                                                    &entry->reg_id);
1109                         if (err) {
1110                                 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1111                                        ha->addr, priv->port, err);
1112                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1113                                 kfree(entry);
1114                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1115                                 break;
1116                         } else {
1117                                 unsigned int mac_hash;
1118                                 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1119                                        ha->addr, priv->port);
1120                                 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1121                                 bucket = &priv->mac_hash[mac_hash];
1122                                 hlist_add_head_rcu(&entry->hlist, bucket);
1123                         }
1124                 }
1125         }
1126
1127         if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1128                 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1129                         priv->port);
1130         } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1131                 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1132                         priv->port);
1133         }
1134 }
1135
1136 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1137 {
1138         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1139                                                  rx_mode_task);
1140         struct mlx4_en_dev *mdev = priv->mdev;
1141         struct net_device *dev = priv->dev;
1142
1143         mutex_lock(&mdev->state_lock);
1144         if (!mdev->device_up) {
1145                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1146                 goto out;
1147         }
1148         if (!priv->port_up) {
1149                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1150                 goto out;
1151         }
1152
1153         if (!netif_carrier_ok(dev)) {
1154                 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1155                         if (priv->port_state.link_state) {
1156                                 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1157                                 netif_carrier_on(dev);
1158                                 en_dbg(LINK, priv, "Link Up\n");
1159                         }
1160                 }
1161         }
1162
1163         if (dev->priv_flags & IFF_UNICAST_FLT)
1164                 mlx4_en_do_uc_filter(priv, dev, mdev);
1165
1166         /* Promsicuous mode: disable all filters */
1167         if ((dev->flags & IFF_PROMISC) ||
1168             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1169                 mlx4_en_set_promisc_mode(priv, mdev);
1170                 goto out;
1171         }
1172
1173         /* Not in promiscuous mode */
1174         if (priv->flags & MLX4_EN_FLAG_PROMISC)
1175                 mlx4_en_clear_promisc_mode(priv, mdev);
1176
1177         mlx4_en_do_multicast(priv, dev, mdev);
1178 out:
1179         mutex_unlock(&mdev->state_lock);
1180 }
1181
1182 #ifdef CONFIG_NET_POLL_CONTROLLER
1183 static void mlx4_en_netpoll(struct net_device *dev)
1184 {
1185         struct mlx4_en_priv *priv = netdev_priv(dev);
1186         struct mlx4_en_cq *cq;
1187         unsigned long flags;
1188         int i;
1189
1190         for (i = 0; i < priv->rx_ring_num; i++) {
1191                 cq = &priv->rx_cq[i];
1192                 spin_lock_irqsave(&cq->lock, flags);
1193                 napi_synchronize(&cq->napi);
1194                 mlx4_en_process_rx_cq(dev, cq, 0);
1195                 spin_unlock_irqrestore(&cq->lock, flags);
1196         }
1197 }
1198 #endif
1199
1200 static void mlx4_en_tx_timeout(struct net_device *dev)
1201 {
1202         struct mlx4_en_priv *priv = netdev_priv(dev);
1203         struct mlx4_en_dev *mdev = priv->mdev;
1204
1205         if (netif_msg_timer(priv))
1206                 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1207
1208         priv->port_stats.tx_timeout++;
1209         en_dbg(DRV, priv, "Scheduling watchdog\n");
1210         queue_work(mdev->workqueue, &priv->watchdog_task);
1211 }
1212
1213
1214 static struct net_device_stats *mlx4_en_get_stats(struct net_device *dev)
1215 {
1216         struct mlx4_en_priv *priv = netdev_priv(dev);
1217
1218         spin_lock_bh(&priv->stats_lock);
1219         memcpy(&priv->ret_stats, &priv->stats, sizeof(priv->stats));
1220         spin_unlock_bh(&priv->stats_lock);
1221
1222         return &priv->ret_stats;
1223 }
1224
1225 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1226 {
1227         struct mlx4_en_cq *cq;
1228         int i;
1229
1230         /* If we haven't received a specific coalescing setting
1231          * (module param), we set the moderation parameters as follows:
1232          * - moder_cnt is set to the number of mtu sized packets to
1233          *   satisfy our coalescing target.
1234          * - moder_time is set to a fixed value.
1235          */
1236         priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1237         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1238         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1239         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1240         en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1241                priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1242
1243         /* Setup cq moderation params */
1244         for (i = 0; i < priv->rx_ring_num; i++) {
1245                 cq = &priv->rx_cq[i];
1246                 cq->moder_cnt = priv->rx_frames;
1247                 cq->moder_time = priv->rx_usecs;
1248                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1249                 priv->last_moder_packets[i] = 0;
1250                 priv->last_moder_bytes[i] = 0;
1251         }
1252
1253         for (i = 0; i < priv->tx_ring_num; i++) {
1254                 cq = &priv->tx_cq[i];
1255                 cq->moder_cnt = priv->tx_frames;
1256                 cq->moder_time = priv->tx_usecs;
1257         }
1258
1259         /* Reset auto-moderation params */
1260         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1261         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1262         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1263         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1264         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1265         priv->adaptive_rx_coal = 1;
1266         priv->last_moder_jiffies = 0;
1267         priv->last_moder_tx_packets = 0;
1268 }
1269
1270 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1271 {
1272         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1273         struct mlx4_en_cq *cq;
1274         unsigned long packets;
1275         unsigned long rate;
1276         unsigned long avg_pkt_size;
1277         unsigned long rx_packets;
1278         unsigned long rx_bytes;
1279         unsigned long rx_pkt_diff;
1280         int moder_time;
1281         int ring, err;
1282
1283         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1284                 return;
1285
1286         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1287                 spin_lock_bh(&priv->stats_lock);
1288                 rx_packets = priv->rx_ring[ring].packets;
1289                 rx_bytes = priv->rx_ring[ring].bytes;
1290                 spin_unlock_bh(&priv->stats_lock);
1291
1292                 rx_pkt_diff = ((unsigned long) (rx_packets -
1293                                 priv->last_moder_packets[ring]));
1294                 packets = rx_pkt_diff;
1295                 rate = packets * HZ / period;
1296                 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1297                                 priv->last_moder_bytes[ring])) / packets : 0;
1298
1299                 /* Apply auto-moderation only when packet rate
1300                  * exceeds a rate that it matters */
1301                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1302                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1303                         if (rate < priv->pkt_rate_low)
1304                                 moder_time = priv->rx_usecs_low;
1305                         else if (rate > priv->pkt_rate_high)
1306                                 moder_time = priv->rx_usecs_high;
1307                         else
1308                                 moder_time = (rate - priv->pkt_rate_low) *
1309                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1310                                         (priv->pkt_rate_high - priv->pkt_rate_low) +
1311                                         priv->rx_usecs_low;
1312                 } else {
1313                         moder_time = priv->rx_usecs_low;
1314                 }
1315
1316                 if (moder_time != priv->last_moder_time[ring]) {
1317                         priv->last_moder_time[ring] = moder_time;
1318                         cq = &priv->rx_cq[ring];
1319                         cq->moder_time = moder_time;
1320                         err = mlx4_en_set_cq_moder(priv, cq);
1321                         if (err)
1322                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1323                                        ring);
1324                 }
1325                 priv->last_moder_packets[ring] = rx_packets;
1326                 priv->last_moder_bytes[ring] = rx_bytes;
1327         }
1328
1329         priv->last_moder_jiffies = jiffies;
1330 }
1331
1332 static void mlx4_en_do_get_stats(struct work_struct *work)
1333 {
1334         struct delayed_work *delay = to_delayed_work(work);
1335         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1336                                                  stats_task);
1337         struct mlx4_en_dev *mdev = priv->mdev;
1338         int err;
1339
1340         mutex_lock(&mdev->state_lock);
1341         if (mdev->device_up) {
1342                 err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1343                 if (err)
1344                         en_dbg(HW, priv, "Could not update stats\n");
1345
1346                 if (priv->port_up)
1347                         mlx4_en_auto_moderation(priv);
1348
1349                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1350         }
1351         if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1352                 queue_work(mdev->workqueue, &priv->mac_task);
1353                 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1354         }
1355         mutex_unlock(&mdev->state_lock);
1356 }
1357
1358 static void mlx4_en_linkstate(struct work_struct *work)
1359 {
1360         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1361                                                  linkstate_task);
1362         struct mlx4_en_dev *mdev = priv->mdev;
1363         int linkstate = priv->link_state;
1364
1365         mutex_lock(&mdev->state_lock);
1366         /* If observable port state changed set carrier state and
1367          * report to system log */
1368         if (priv->last_link_state != linkstate) {
1369                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1370                         en_info(priv, "Link Down\n");
1371                         netif_carrier_off(priv->dev);
1372                 } else {
1373                         en_info(priv, "Link Up\n");
1374                         netif_carrier_on(priv->dev);
1375                 }
1376         }
1377         priv->last_link_state = linkstate;
1378         mutex_unlock(&mdev->state_lock);
1379 }
1380
1381
1382 int mlx4_en_start_port(struct net_device *dev)
1383 {
1384         struct mlx4_en_priv *priv = netdev_priv(dev);
1385         struct mlx4_en_dev *mdev = priv->mdev;
1386         struct mlx4_en_cq *cq;
1387         struct mlx4_en_tx_ring *tx_ring;
1388         int rx_index = 0;
1389         int tx_index = 0;
1390         int err = 0;
1391         int i;
1392         int j;
1393         u8 mc_list[16] = {0};
1394
1395         if (priv->port_up) {
1396                 en_dbg(DRV, priv, "start port called while port already up\n");
1397                 return 0;
1398         }
1399
1400         INIT_LIST_HEAD(&priv->mc_list);
1401         INIT_LIST_HEAD(&priv->curr_list);
1402         INIT_LIST_HEAD(&priv->ethtool_list);
1403         memset(&priv->ethtool_rules[0], 0,
1404                sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1405
1406         /* Calculate Rx buf size */
1407         dev->mtu = min(dev->mtu, priv->max_mtu);
1408         mlx4_en_calc_rx_buf(dev);
1409         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1410
1411         /* Configure rx cq's and rings */
1412         err = mlx4_en_activate_rx_rings(priv);
1413         if (err) {
1414                 en_err(priv, "Failed to activate RX rings\n");
1415                 return err;
1416         }
1417         for (i = 0; i < priv->rx_ring_num; i++) {
1418                 cq = &priv->rx_cq[i];
1419
1420                 err = mlx4_en_activate_cq(priv, cq, i);
1421                 if (err) {
1422                         en_err(priv, "Failed activating Rx CQ\n");
1423                         goto cq_err;
1424                 }
1425                 for (j = 0; j < cq->size; j++)
1426                         cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1427                 err = mlx4_en_set_cq_moder(priv, cq);
1428                 if (err) {
1429                         en_err(priv, "Failed setting cq moderation parameters");
1430                         mlx4_en_deactivate_cq(priv, cq);
1431                         goto cq_err;
1432                 }
1433                 mlx4_en_arm_cq(priv, cq);
1434                 priv->rx_ring[i].cqn = cq->mcq.cqn;
1435                 ++rx_index;
1436         }
1437
1438         /* Set qp number */
1439         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1440         err = mlx4_en_get_qp(priv);
1441         if (err) {
1442                 en_err(priv, "Failed getting eth qp\n");
1443                 goto cq_err;
1444         }
1445         mdev->mac_removed[priv->port] = 0;
1446
1447         err = mlx4_en_config_rss_steer(priv);
1448         if (err) {
1449                 en_err(priv, "Failed configuring rss steering\n");
1450                 goto mac_err;
1451         }
1452
1453         err = mlx4_en_create_drop_qp(priv);
1454         if (err)
1455                 goto rss_err;
1456
1457         /* Configure tx cq's and rings */
1458         for (i = 0; i < priv->tx_ring_num; i++) {
1459                 /* Configure cq */
1460                 cq = &priv->tx_cq[i];
1461                 err = mlx4_en_activate_cq(priv, cq, i);
1462                 if (err) {
1463                         en_err(priv, "Failed allocating Tx CQ\n");
1464                         goto tx_err;
1465                 }
1466                 err = mlx4_en_set_cq_moder(priv, cq);
1467                 if (err) {
1468                         en_err(priv, "Failed setting cq moderation parameters");
1469                         mlx4_en_deactivate_cq(priv, cq);
1470                         goto tx_err;
1471                 }
1472                 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1473                 cq->buf->wqe_index = cpu_to_be16(0xffff);
1474
1475                 /* Configure ring */
1476                 tx_ring = &priv->tx_ring[i];
1477                 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1478                         i / priv->num_tx_rings_p_up);
1479                 if (err) {
1480                         en_err(priv, "Failed allocating Tx ring\n");
1481                         mlx4_en_deactivate_cq(priv, cq);
1482                         goto tx_err;
1483                 }
1484                 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1485
1486                 /* Arm CQ for TX completions */
1487                 mlx4_en_arm_cq(priv, cq);
1488
1489                 /* Set initial ownership of all Tx TXBBs to SW (1) */
1490                 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1491                         *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1492                 ++tx_index;
1493         }
1494
1495         /* Configure port */
1496         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1497                                     priv->rx_skb_size + ETH_FCS_LEN,
1498                                     priv->prof->tx_pause,
1499                                     priv->prof->tx_ppp,
1500                                     priv->prof->rx_pause,
1501                                     priv->prof->rx_ppp);
1502         if (err) {
1503                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1504                        priv->port, err);
1505                 goto tx_err;
1506         }
1507         /* Set default qp number */
1508         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1509         if (err) {
1510                 en_err(priv, "Failed setting default qp numbers\n");
1511                 goto tx_err;
1512         }
1513
1514         /* Init port */
1515         en_dbg(HW, priv, "Initializing port\n");
1516         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1517         if (err) {
1518                 en_err(priv, "Failed Initializing port\n");
1519                 goto tx_err;
1520         }
1521
1522         /* Attach rx QP to bradcast address */
1523         memset(&mc_list[10], 0xff, ETH_ALEN);
1524         mc_list[5] = priv->port; /* needed for B0 steering support */
1525         if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1526                                   priv->port, 0, MLX4_PROT_ETH,
1527                                   &priv->broadcast_id))
1528                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1529
1530         /* Must redo promiscuous mode setup. */
1531         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1532
1533         /* Schedule multicast task to populate multicast list */
1534         queue_work(mdev->workqueue, &priv->rx_mode_task);
1535
1536         mlx4_set_stats_bitmap(mdev->dev, &priv->stats_bitmap);
1537
1538         priv->port_up = true;
1539         netif_tx_start_all_queues(dev);
1540         netif_device_attach(dev);
1541
1542         return 0;
1543
1544 tx_err:
1545         while (tx_index--) {
1546                 mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[tx_index]);
1547                 mlx4_en_deactivate_cq(priv, &priv->tx_cq[tx_index]);
1548         }
1549         mlx4_en_destroy_drop_qp(priv);
1550 rss_err:
1551         mlx4_en_release_rss_steer(priv);
1552 mac_err:
1553         mlx4_en_put_qp(priv);
1554 cq_err:
1555         while (rx_index--)
1556                 mlx4_en_deactivate_cq(priv, &priv->rx_cq[rx_index]);
1557         for (i = 0; i < priv->rx_ring_num; i++)
1558                 mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
1559
1560         return err; /* need to close devices */
1561 }
1562
1563
1564 void mlx4_en_stop_port(struct net_device *dev, int detach)
1565 {
1566         struct mlx4_en_priv *priv = netdev_priv(dev);
1567         struct mlx4_en_dev *mdev = priv->mdev;
1568         struct mlx4_en_mc_list *mclist, *tmp;
1569         struct ethtool_flow_id *flow, *tmp_flow;
1570         int i;
1571         u8 mc_list[16] = {0};
1572
1573         if (!priv->port_up) {
1574                 en_dbg(DRV, priv, "stop port called while port already down\n");
1575                 return;
1576         }
1577
1578         /* Synchronize with tx routine */
1579         netif_tx_lock_bh(dev);
1580         if (detach)
1581                 netif_device_detach(dev);
1582         netif_tx_stop_all_queues(dev);
1583         netif_tx_unlock_bh(dev);
1584
1585         netif_tx_disable(dev);
1586
1587         /* Set port as not active */
1588         priv->port_up = false;
1589
1590         /* Promsicuous mode */
1591         if (mdev->dev->caps.steering_mode ==
1592             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1593                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1594                                  MLX4_EN_FLAG_MC_PROMISC);
1595                 mlx4_flow_steer_promisc_remove(mdev->dev,
1596                                                priv->port,
1597                                                MLX4_FS_PROMISC_UPLINK);
1598                 mlx4_flow_steer_promisc_remove(mdev->dev,
1599                                                priv->port,
1600                                                MLX4_FS_PROMISC_ALL_MULTI);
1601         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1602                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1603
1604                 /* Disable promiscouos mode */
1605                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1606                                             priv->port);
1607
1608                 /* Disable Multicast promisc */
1609                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1610                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1611                                                       priv->port);
1612                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1613                 }
1614         }
1615
1616         /* Detach All multicasts */
1617         memset(&mc_list[10], 0xff, ETH_ALEN);
1618         mc_list[5] = priv->port; /* needed for B0 steering support */
1619         mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1620                               MLX4_PROT_ETH, priv->broadcast_id);
1621         list_for_each_entry(mclist, &priv->curr_list, list) {
1622                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1623                 mc_list[5] = priv->port;
1624                 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1625                                       mc_list, MLX4_PROT_ETH, mclist->reg_id);
1626         }
1627         mlx4_en_clear_list(dev);
1628         list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1629                 list_del(&mclist->list);
1630                 kfree(mclist);
1631         }
1632
1633         /* Flush multicast filter */
1634         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1635
1636         mlx4_en_destroy_drop_qp(priv);
1637
1638         /* Free TX Rings */
1639         for (i = 0; i < priv->tx_ring_num; i++) {
1640                 mlx4_en_deactivate_tx_ring(priv, &priv->tx_ring[i]);
1641                 mlx4_en_deactivate_cq(priv, &priv->tx_cq[i]);
1642         }
1643         msleep(10);
1644
1645         for (i = 0; i < priv->tx_ring_num; i++)
1646                 mlx4_en_free_tx_buf(dev, &priv->tx_ring[i]);
1647
1648         /* Free RSS qps */
1649         mlx4_en_release_rss_steer(priv);
1650
1651         /* Unregister Mac address for the port */
1652         mlx4_en_put_qp(priv);
1653         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAGS2_REASSIGN_MAC_EN))
1654                 mdev->mac_removed[priv->port] = 1;
1655
1656         /* Remove flow steering rules for the port*/
1657         if (mdev->dev->caps.steering_mode ==
1658             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1659                 ASSERT_RTNL();
1660                 list_for_each_entry_safe(flow, tmp_flow,
1661                                          &priv->ethtool_list, list) {
1662                         mlx4_flow_detach(mdev->dev, flow->id);
1663                         list_del(&flow->list);
1664                 }
1665         }
1666
1667         /* Free RX Rings */
1668         for (i = 0; i < priv->rx_ring_num; i++) {
1669                 mlx4_en_deactivate_rx_ring(priv, &priv->rx_ring[i]);
1670                 while (test_bit(NAPI_STATE_SCHED, &priv->rx_cq[i].napi.state))
1671                         msleep(1);
1672                 mlx4_en_deactivate_cq(priv, &priv->rx_cq[i]);
1673         }
1674
1675         /* close port*/
1676         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1677 }
1678
1679 static void mlx4_en_restart(struct work_struct *work)
1680 {
1681         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1682                                                  watchdog_task);
1683         struct mlx4_en_dev *mdev = priv->mdev;
1684         struct net_device *dev = priv->dev;
1685
1686         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1687
1688         mutex_lock(&mdev->state_lock);
1689         if (priv->port_up) {
1690                 mlx4_en_stop_port(dev, 1);
1691                 if (mlx4_en_start_port(dev))
1692                         en_err(priv, "Failed restarting port %d\n", priv->port);
1693         }
1694         mutex_unlock(&mdev->state_lock);
1695 }
1696
1697 static void mlx4_en_clear_stats(struct net_device *dev)
1698 {
1699         struct mlx4_en_priv *priv = netdev_priv(dev);
1700         struct mlx4_en_dev *mdev = priv->mdev;
1701         int i;
1702
1703         if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1704                 en_dbg(HW, priv, "Failed dumping statistics\n");
1705
1706         memset(&priv->stats, 0, sizeof(priv->stats));
1707         memset(&priv->pstats, 0, sizeof(priv->pstats));
1708         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1709         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1710
1711         for (i = 0; i < priv->tx_ring_num; i++) {
1712                 priv->tx_ring[i].bytes = 0;
1713                 priv->tx_ring[i].packets = 0;
1714                 priv->tx_ring[i].tx_csum = 0;
1715         }
1716         for (i = 0; i < priv->rx_ring_num; i++) {
1717                 priv->rx_ring[i].bytes = 0;
1718                 priv->rx_ring[i].packets = 0;
1719                 priv->rx_ring[i].csum_ok = 0;
1720                 priv->rx_ring[i].csum_none = 0;
1721         }
1722 }
1723
1724 static int mlx4_en_open(struct net_device *dev)
1725 {
1726         struct mlx4_en_priv *priv = netdev_priv(dev);
1727         struct mlx4_en_dev *mdev = priv->mdev;
1728         int err = 0;
1729
1730         mutex_lock(&mdev->state_lock);
1731
1732         if (!mdev->device_up) {
1733                 en_err(priv, "Cannot open - device down/disabled\n");
1734                 err = -EBUSY;
1735                 goto out;
1736         }
1737
1738         /* Reset HW statistics and SW counters */
1739         mlx4_en_clear_stats(dev);
1740
1741         err = mlx4_en_start_port(dev);
1742         if (err)
1743                 en_err(priv, "Failed starting port:%d\n", priv->port);
1744
1745 out:
1746         mutex_unlock(&mdev->state_lock);
1747         return err;
1748 }
1749
1750
1751 static int mlx4_en_close(struct net_device *dev)
1752 {
1753         struct mlx4_en_priv *priv = netdev_priv(dev);
1754         struct mlx4_en_dev *mdev = priv->mdev;
1755
1756         en_dbg(IFDOWN, priv, "Close port called\n");
1757
1758         mutex_lock(&mdev->state_lock);
1759
1760         mlx4_en_stop_port(dev, 0);
1761         netif_carrier_off(dev);
1762
1763         mutex_unlock(&mdev->state_lock);
1764         return 0;
1765 }
1766
1767 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1768 {
1769         int i;
1770
1771 #ifdef CONFIG_RFS_ACCEL
1772         free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1773         priv->dev->rx_cpu_rmap = NULL;
1774 #endif
1775
1776         for (i = 0; i < priv->tx_ring_num; i++) {
1777                 if (priv->tx_ring[i].tx_info)
1778                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1779                 if (priv->tx_cq[i].buf)
1780                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1781         }
1782
1783         for (i = 0; i < priv->rx_ring_num; i++) {
1784                 if (priv->rx_ring[i].rx_info)
1785                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1786                                 priv->prof->rx_ring_size, priv->stride);
1787                 if (priv->rx_cq[i].buf)
1788                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1789         }
1790
1791         if (priv->base_tx_qpn) {
1792                 mlx4_qp_release_range(priv->mdev->dev, priv->base_tx_qpn, priv->tx_ring_num);
1793                 priv->base_tx_qpn = 0;
1794         }
1795 }
1796
1797 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1798 {
1799         struct mlx4_en_port_profile *prof = priv->prof;
1800         int i;
1801         int err;
1802
1803         err = mlx4_qp_reserve_range(priv->mdev->dev, priv->tx_ring_num, 256, &priv->base_tx_qpn);
1804         if (err) {
1805                 en_err(priv, "failed reserving range for TX rings\n");
1806                 return err;
1807         }
1808
1809         /* Create tx Rings */
1810         for (i = 0; i < priv->tx_ring_num; i++) {
1811                 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1812                                       prof->tx_ring_size, i, TX))
1813                         goto err;
1814
1815                 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i], priv->base_tx_qpn + i,
1816                                            prof->tx_ring_size, TXBB_SIZE))
1817                         goto err;
1818         }
1819
1820         /* Create rx Rings */
1821         for (i = 0; i < priv->rx_ring_num; i++) {
1822                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
1823                                       prof->rx_ring_size, i, RX))
1824                         goto err;
1825
1826                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
1827                                            prof->rx_ring_size, priv->stride))
1828                         goto err;
1829         }
1830
1831 #ifdef CONFIG_RFS_ACCEL
1832         priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->mdev->dev->caps.comp_pool);
1833         if (!priv->dev->rx_cpu_rmap)
1834                 goto err;
1835 #endif
1836
1837         return 0;
1838
1839 err:
1840         en_err(priv, "Failed to allocate NIC resources\n");
1841         return -ENOMEM;
1842 }
1843
1844
1845 void mlx4_en_destroy_netdev(struct net_device *dev)
1846 {
1847         struct mlx4_en_priv *priv = netdev_priv(dev);
1848         struct mlx4_en_dev *mdev = priv->mdev;
1849
1850         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
1851
1852         /* Unregister device - this will close the port if it was up */
1853         if (priv->registered)
1854                 unregister_netdev(dev);
1855
1856         if (priv->allocated)
1857                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
1858
1859         cancel_delayed_work(&priv->stats_task);
1860         /* flush any pending task for this netdev */
1861         flush_workqueue(mdev->workqueue);
1862
1863         /* Detach the netdev so tasks would not attempt to access it */
1864         mutex_lock(&mdev->state_lock);
1865         mdev->pndev[priv->port] = NULL;
1866         mutex_unlock(&mdev->state_lock);
1867
1868         mlx4_en_free_resources(priv);
1869
1870         kfree(priv->tx_ring);
1871         kfree(priv->tx_cq);
1872
1873         free_netdev(dev);
1874 }
1875
1876 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
1877 {
1878         struct mlx4_en_priv *priv = netdev_priv(dev);
1879         struct mlx4_en_dev *mdev = priv->mdev;
1880         int err = 0;
1881
1882         en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
1883                  dev->mtu, new_mtu);
1884
1885         if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
1886                 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
1887                 return -EPERM;
1888         }
1889         dev->mtu = new_mtu;
1890
1891         if (netif_running(dev)) {
1892                 mutex_lock(&mdev->state_lock);
1893                 if (!mdev->device_up) {
1894                         /* NIC is probably restarting - let watchdog task reset
1895                          * the port */
1896                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
1897                 } else {
1898                         mlx4_en_stop_port(dev, 1);
1899                         err = mlx4_en_start_port(dev);
1900                         if (err) {
1901                                 en_err(priv, "Failed restarting port:%d\n",
1902                                          priv->port);
1903                                 queue_work(mdev->workqueue, &priv->watchdog_task);
1904                         }
1905                 }
1906                 mutex_unlock(&mdev->state_lock);
1907         }
1908         return 0;
1909 }
1910
1911 static int mlx4_en_set_features(struct net_device *netdev,
1912                 netdev_features_t features)
1913 {
1914         struct mlx4_en_priv *priv = netdev_priv(netdev);
1915
1916         if (features & NETIF_F_LOOPBACK)
1917                 priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
1918         else
1919                 priv->ctrl_flags &=
1920                         cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
1921
1922         mlx4_en_update_loopback_state(netdev, features);
1923
1924         return 0;
1925
1926 }
1927
1928 static int mlx4_en_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1929                            struct net_device *dev,
1930                            const unsigned char *addr, u16 flags)
1931 {
1932         struct mlx4_en_priv *priv = netdev_priv(dev);
1933         struct mlx4_dev *mdev = priv->mdev->dev;
1934         int err;
1935
1936         if (!mlx4_is_mfunc(mdev))
1937                 return -EOPNOTSUPP;
1938
1939         /* Hardware does not support aging addresses, allow only
1940          * permanent addresses if ndm_state is given
1941          */
1942         if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
1943                 en_info(priv, "Add FDB only supports static addresses\n");
1944                 return -EINVAL;
1945         }
1946
1947         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr))
1948                 err = dev_uc_add_excl(dev, addr);
1949         else if (is_multicast_ether_addr(addr))
1950                 err = dev_mc_add_excl(dev, addr);
1951         else
1952                 err = -EINVAL;
1953
1954         /* Only return duplicate errors if NLM_F_EXCL is set */
1955         if (err == -EEXIST && !(flags & NLM_F_EXCL))
1956                 err = 0;
1957
1958         return err;
1959 }
1960
1961 static int mlx4_en_fdb_del(struct ndmsg *ndm,
1962                            struct nlattr *tb[],
1963                            struct net_device *dev,
1964                            const unsigned char *addr)
1965 {
1966         struct mlx4_en_priv *priv = netdev_priv(dev);
1967         struct mlx4_dev *mdev = priv->mdev->dev;
1968         int err;
1969
1970         if (!mlx4_is_mfunc(mdev))
1971                 return -EOPNOTSUPP;
1972
1973         if (ndm->ndm_state && !(ndm->ndm_state & NUD_PERMANENT)) {
1974                 en_info(priv, "Del FDB only supports static addresses\n");
1975                 return -EINVAL;
1976         }
1977
1978         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr))
1979                 err = dev_uc_del(dev, addr);
1980         else if (is_multicast_ether_addr(addr))
1981                 err = dev_mc_del(dev, addr);
1982         else
1983                 err = -EINVAL;
1984
1985         return err;
1986 }
1987
1988 static int mlx4_en_fdb_dump(struct sk_buff *skb,
1989                             struct netlink_callback *cb,
1990                             struct net_device *dev, int idx)
1991 {
1992         struct mlx4_en_priv *priv = netdev_priv(dev);
1993         struct mlx4_dev *mdev = priv->mdev->dev;
1994
1995         if (mlx4_is_mfunc(mdev))
1996                 idx = ndo_dflt_fdb_dump(skb, cb, dev, idx);
1997
1998         return idx;
1999 }
2000
2001 static const struct net_device_ops mlx4_netdev_ops = {
2002         .ndo_open               = mlx4_en_open,
2003         .ndo_stop               = mlx4_en_close,
2004         .ndo_start_xmit         = mlx4_en_xmit,
2005         .ndo_select_queue       = mlx4_en_select_queue,
2006         .ndo_get_stats          = mlx4_en_get_stats,
2007         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2008         .ndo_set_mac_address    = mlx4_en_set_mac,
2009         .ndo_validate_addr      = eth_validate_addr,
2010         .ndo_change_mtu         = mlx4_en_change_mtu,
2011         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2012         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2013         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2014 #ifdef CONFIG_NET_POLL_CONTROLLER
2015         .ndo_poll_controller    = mlx4_en_netpoll,
2016 #endif
2017         .ndo_set_features       = mlx4_en_set_features,
2018         .ndo_setup_tc           = mlx4_en_setup_tc,
2019 #ifdef CONFIG_RFS_ACCEL
2020         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2021 #endif
2022         .ndo_fdb_add            = mlx4_en_fdb_add,
2023         .ndo_fdb_del            = mlx4_en_fdb_del,
2024         .ndo_fdb_dump           = mlx4_en_fdb_dump,
2025 };
2026
2027 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
2028                         struct mlx4_en_port_profile *prof)
2029 {
2030         struct net_device *dev;
2031         struct mlx4_en_priv *priv;
2032         int i;
2033         int err;
2034
2035         dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
2036                                  MAX_TX_RINGS, MAX_RX_RINGS);
2037         if (dev == NULL)
2038                 return -ENOMEM;
2039
2040         netif_set_real_num_tx_queues(dev, prof->tx_ring_num);
2041         netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
2042
2043         SET_NETDEV_DEV(dev, &mdev->dev->pdev->dev);
2044         dev->dev_id =  port - 1;
2045
2046         /*
2047          * Initialize driver private data
2048          */
2049
2050         priv = netdev_priv(dev);
2051         memset(priv, 0, sizeof(struct mlx4_en_priv));
2052         priv->dev = dev;
2053         priv->mdev = mdev;
2054         priv->ddev = &mdev->pdev->dev;
2055         priv->prof = prof;
2056         priv->port = port;
2057         priv->port_up = false;
2058         priv->flags = prof->flags;
2059         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2060                         MLX4_WQE_CTRL_SOLICITED);
2061         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2062         priv->tx_ring_num = prof->tx_ring_num;
2063
2064         priv->tx_ring = kzalloc(sizeof(struct mlx4_en_tx_ring) * MAX_TX_RINGS,
2065                                 GFP_KERNEL);
2066         if (!priv->tx_ring) {
2067                 err = -ENOMEM;
2068                 goto out;
2069         }
2070         priv->tx_cq = kzalloc(sizeof(struct mlx4_en_cq) * MAX_TX_RINGS,
2071                               GFP_KERNEL);
2072         if (!priv->tx_cq) {
2073                 err = -ENOMEM;
2074                 goto out;
2075         }
2076         priv->rx_ring_num = prof->rx_ring_num;
2077         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2078         priv->mac_index = -1;
2079         priv->msg_enable = MLX4_EN_MSG_LEVEL;
2080         spin_lock_init(&priv->stats_lock);
2081         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2082         INIT_WORK(&priv->mac_task, mlx4_en_do_set_mac);
2083         INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2084         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2085         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2086 #ifdef CONFIG_MLX4_EN_DCB
2087         if (!mlx4_is_slave(priv->mdev->dev))
2088                 dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2089 #endif
2090
2091         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2092                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2093
2094         /* Query for default mac and max mtu */
2095         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2096
2097         /* Set default MAC */
2098         dev->addr_len = ETH_ALEN;
2099         mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
2100         if (!is_valid_ether_addr(dev->dev_addr)) {
2101                 en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
2102                        priv->port, dev->dev_addr);
2103                 err = -EINVAL;
2104                 goto out;
2105         }
2106
2107         memcpy(priv->prev_mac, dev->dev_addr, sizeof(priv->prev_mac));
2108
2109         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2110                                           DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2111         err = mlx4_en_alloc_resources(priv);
2112         if (err)
2113                 goto out;
2114
2115 #ifdef CONFIG_RFS_ACCEL
2116         INIT_LIST_HEAD(&priv->filters);
2117         spin_lock_init(&priv->filters_lock);
2118 #endif
2119
2120         /* Allocate page for receive rings */
2121         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2122                                 MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2123         if (err) {
2124                 en_err(priv, "Failed to allocate page for rx qps\n");
2125                 goto out;
2126         }
2127         priv->allocated = 1;
2128
2129         /*
2130          * Initialize netdev entry points
2131          */
2132         dev->netdev_ops = &mlx4_netdev_ops;
2133         dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
2134         netif_set_real_num_tx_queues(dev, priv->tx_ring_num);
2135         netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
2136
2137         SET_ETHTOOL_OPS(dev, &mlx4_en_ethtool_ops);
2138
2139         /*
2140          * Set driver features
2141          */
2142         dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2143         if (mdev->LSO_support)
2144                 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2145
2146         dev->vlan_features = dev->hw_features;
2147
2148         dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
2149         dev->features = dev->hw_features | NETIF_F_HIGHDMA |
2150                         NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX |
2151                         NETIF_F_HW_VLAN_FILTER;
2152         dev->hw_features |= NETIF_F_LOOPBACK;
2153
2154         if (mdev->dev->caps.steering_mode ==
2155             MLX4_STEERING_MODE_DEVICE_MANAGED)
2156                 dev->hw_features |= NETIF_F_NTUPLE;
2157
2158         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
2159                 dev->priv_flags |= IFF_UNICAST_FLT;
2160
2161         mdev->pndev[port] = dev;
2162
2163         netif_carrier_off(dev);
2164         err = register_netdev(dev);
2165         if (err) {
2166                 en_err(priv, "Netdev registration failed for port %d\n", port);
2167                 goto out;
2168         }
2169         priv->registered = 1;
2170
2171         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2172         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2173
2174         mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
2175
2176         /* Configure port */
2177         mlx4_en_calc_rx_buf(dev);
2178         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2179                                     priv->rx_skb_size + ETH_FCS_LEN,
2180                                     prof->tx_pause, prof->tx_ppp,
2181                                     prof->rx_pause, prof->rx_ppp);
2182         if (err) {
2183                 en_err(priv, "Failed setting port general configurations "
2184                        "for port %d, with error %d\n", priv->port, err);
2185                 goto out;
2186         }
2187
2188         /* Init port */
2189         en_warn(priv, "Initializing port\n");
2190         err = mlx4_INIT_PORT(mdev->dev, priv->port);
2191         if (err) {
2192                 en_err(priv, "Failed Initializing port\n");
2193                 goto out;
2194         }
2195         mlx4_en_set_default_moderation(priv);
2196         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2197         return 0;
2198
2199 out:
2200         mlx4_en_destroy_netdev(dev);
2201         return err;
2202 }
2203