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