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/bpf.h>
  35 #include <linux/etherdevice.h>
  36 #include <linux/tcp.h>
  37 #include <linux/if_vlan.h>
  38 #include <linux/delay.h>
  39 #include <linux/slab.h>
  40 #include <linux/hash.h>
  41 #include <net/ip.h>
  42 #include <net/busy_poll.h>
  43 #include <net/vxlan.h>
  44 #include <net/devlink.h>
  45
  46 #include <linux/mlx4/driver.h>
  47 #include <linux/mlx4/device.h>
  48 #include <linux/mlx4/cmd.h>
  49 #include <linux/mlx4/cq.h>
  50
  51 #include "mlx4_en.h"
  52 #include "en_port.h"
  53
  54 #define MLX4_EN_MAX_XDP_MTU ((int)(PAGE_SIZE - ETH_HLEN - (2 * VLAN_HLEN) - \
  55                                    XDP_PACKET_HEADROOM))
  56
  57 int mlx4_en_setup_tc(struct net_device *dev, u8 up)
  58 {
  59         struct mlx4_en_priv *priv = netdev_priv(dev);
  60         int i;
  61         unsigned int offset = 0;
  62
  63         if (up && up != MLX4_EN_NUM_UP_HIGH)
  64                 return -EINVAL;
  65
  66         netdev_set_num_tc(dev, up);
  67         netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
  68         /* Partition Tx queues evenly amongst UP's */
  69         for (i = 0; i < up; i++) {
  70                 netdev_set_tc_queue(dev, i, priv->num_tx_rings_p_up, offset);
  71                 offset += priv->num_tx_rings_p_up;
  72         }
  73
  74 #ifdef CONFIG_MLX4_EN_DCB
  75         if (!mlx4_is_slave(priv->mdev->dev)) {
  76                 if (up) {
  77                         if (priv->dcbx_cap)
  78                                 priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
  79                 } else {
  80                         priv->flags &= ~MLX4_EN_FLAG_DCB_ENABLED;
  81                         priv->cee_config.pfc_state = false;
  82                 }
  83         }
  84 #endif /* CONFIG_MLX4_EN_DCB */
  85
  86         return 0;
  87 }
  88
  89 int mlx4_en_alloc_tx_queue_per_tc(struct net_device *dev, u8 tc)
  90 {
  91         struct mlx4_en_priv *priv = netdev_priv(dev);
  92         struct mlx4_en_dev *mdev = priv->mdev;
  93         struct mlx4_en_port_profile new_prof;
  94         struct mlx4_en_priv *tmp;
  95         int port_up = 0;
  96         int err = 0;
  97
  98         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
  99         if (!tmp)
 100                 return -ENOMEM;
 101
 102         mutex_lock(&mdev->state_lock);
 103         memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
 104         new_prof.num_up = (tc == 0) ? MLX4_EN_NUM_UP_LOW :
 105                                       MLX4_EN_NUM_UP_HIGH;
 106         new_prof.tx_ring_num[TX] = new_prof.num_tx_rings_p_up *
 107                                    new_prof.num_up;
 108         err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
 109         if (err)
 110                 goto out;
 111
 112         if (priv->port_up) {
 113                 port_up = 1;
 114                 mlx4_en_stop_port(dev, 1);
 115         }
 116
 117         mlx4_en_safe_replace_resources(priv, tmp);
 118         if (port_up) {
 119                 err = mlx4_en_start_port(dev);
 120                 if (err) {
 121                         en_err(priv, "Failed starting port for setup TC\n");
 122                         goto out;
 123                 }
 124         }
 125
 126         err = mlx4_en_setup_tc(dev, tc);
 127 out:
 128         mutex_unlock(&mdev->state_lock);
 129         kfree(tmp);
 130         return err;
 131 }
 132
 133 static int __mlx4_en_setup_tc(struct net_device *dev, u32 handle,
 134                               u32 chain_index, __be16 proto,
 135                               struct tc_to_netdev *tc)
 136 {
 137         if (tc->type != TC_SETUP_MQPRIO)
 138                 return -EINVAL;
 139
 140         if (tc->mqprio->num_tc && tc->mqprio->num_tc != MLX4_EN_NUM_UP_HIGH)
 141                 return -EINVAL;
 142
 143         tc->mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
 144
 145         return mlx4_en_alloc_tx_queue_per_tc(dev, tc->mqprio->num_tc);
 146 }
 147
 148 #ifdef CONFIG_RFS_ACCEL
 149
 150 struct mlx4_en_filter {
 151         struct list_head next;
 152         struct work_struct work;
 153
 154         u8     ip_proto;
 155         __be32 src_ip;
 156         __be32 dst_ip;
 157         __be16 src_port;
 158         __be16 dst_port;
 159
 160         int rxq_index;
 161         struct mlx4_en_priv *priv;
 162         u32 flow_id;                    /* RFS infrastructure id */
 163         int id;                         /* mlx4_en driver id */
 164         u64 reg_id;                     /* Flow steering API id */
 165         u8 activated;                   /* Used to prevent expiry before filter
 166                                          * is attached
 167                                          */
 168         struct hlist_node filter_chain;
 169 };
 170
 171 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
 172
 173 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
 174 {
 175         switch (ip_proto) {
 176         case IPPROTO_UDP:
 177                 return MLX4_NET_TRANS_RULE_ID_UDP;
 178         case IPPROTO_TCP:
 179                 return MLX4_NET_TRANS_RULE_ID_TCP;
 180         default:
 181                 return MLX4_NET_TRANS_RULE_NUM;
 182         }
 183 };
 184
 185 /* Must not acquire state_lock, as its corresponding work_sync
 186  * is done under it.
 187  */
 188 static void mlx4_en_filter_work(struct work_struct *work)
 189 {
 190         struct mlx4_en_filter *filter = container_of(work,
 191                                                      struct mlx4_en_filter,
 192                                                      work);
 193         struct mlx4_en_priv *priv = filter->priv;
 194         struct mlx4_spec_list spec_tcp_udp = {
 195                 .id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
 196                 {
 197                         .tcp_udp = {
 198                                 .dst_port = filter->dst_port,
 199                                 .dst_port_msk = (__force __be16)-1,
 200                                 .src_port = filter->src_port,
 201                                 .src_port_msk = (__force __be16)-1,
 202                         },
 203                 },
 204         };
 205         struct mlx4_spec_list spec_ip = {
 206                 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
 207                 {
 208                         .ipv4 = {
 209                                 .dst_ip = filter->dst_ip,
 210                                 .dst_ip_msk = (__force __be32)-1,
 211                                 .src_ip = filter->src_ip,
 212                                 .src_ip_msk = (__force __be32)-1,
 213                         },
 214                 },
 215         };
 216         struct mlx4_spec_list spec_eth = {
 217                 .id = MLX4_NET_TRANS_RULE_ID_ETH,
 218         };
 219         struct mlx4_net_trans_rule rule = {
 220                 .list = LIST_HEAD_INIT(rule.list),
 221                 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
 222                 .exclusive = 1,
 223                 .allow_loopback = 1,
 224                 .promisc_mode = MLX4_FS_REGULAR,
 225                 .port = priv->port,
 226                 .priority = MLX4_DOMAIN_RFS,
 227         };
 228         int rc;
 229         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 230
 231         if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
 232                 en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
 233                         filter->ip_proto);
 234                 goto ignore;
 235         }
 236         list_add_tail(&spec_eth.list, &rule.list);
 237         list_add_tail(&spec_ip.list, &rule.list);
 238         list_add_tail(&spec_tcp_udp.list, &rule.list);
 239
 240         rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
 241         memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
 242         memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 243
 244         filter->activated = 0;
 245
 246         if (filter->reg_id) {
 247                 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 248                 if (rc && rc != -ENOENT)
 249                         en_err(priv, "Error detaching flow. rc = %d\n", rc);
 250         }
 251
 252         rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
 253         if (rc)
 254                 en_err(priv, "Error attaching flow. err = %d\n", rc);
 255
 256 ignore:
 257         mlx4_en_filter_rfs_expire(priv);
 258
 259         filter->activated = 1;
 260 }
 261
 262 static inline struct hlist_head *
 263 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 264                    __be16 src_port, __be16 dst_port)
 265 {
 266         unsigned long l;
 267         int bucket_idx;
 268
 269         l = (__force unsigned long)src_port |
 270             ((__force unsigned long)dst_port << 2);
 271         l ^= (__force unsigned long)(src_ip ^ dst_ip);
 272
 273         bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
 274
 275         return &priv->filter_hash[bucket_idx];
 276 }
 277
 278 static struct mlx4_en_filter *
 279 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
 280                      __be32 dst_ip, u8 ip_proto, __be16 src_port,
 281                      __be16 dst_port, u32 flow_id)
 282 {
 283         struct mlx4_en_filter *filter = NULL;
 284
 285         filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
 286         if (!filter)
 287                 return NULL;
 288
 289         filter->priv = priv;
 290         filter->rxq_index = rxq_index;
 291         INIT_WORK(&filter->work, mlx4_en_filter_work);
 292
 293         filter->src_ip = src_ip;
 294         filter->dst_ip = dst_ip;
 295         filter->ip_proto = ip_proto;
 296         filter->src_port = src_port;
 297         filter->dst_port = dst_port;
 298
 299         filter->flow_id = flow_id;
 300
 301         filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
 302
 303         list_add_tail(&filter->next, &priv->filters);
 304         hlist_add_head(&filter->filter_chain,
 305                        filter_hash_bucket(priv, src_ip, dst_ip, src_port,
 306                                           dst_port));
 307
 308         return filter;
 309 }
 310
 311 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
 312 {
 313         struct mlx4_en_priv *priv = filter->priv;
 314         int rc;
 315
 316         list_del(&filter->next);
 317
 318         rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 319         if (rc && rc != -ENOENT)
 320                 en_err(priv, "Error detaching flow. rc = %d\n", rc);
 321
 322         kfree(filter);
 323 }
 324
 325 static inline struct mlx4_en_filter *
 326 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 327                     u8 ip_proto, __be16 src_port, __be16 dst_port)
 328 {
 329         struct mlx4_en_filter *filter;
 330         struct mlx4_en_filter *ret = NULL;
 331
 332         hlist_for_each_entry(filter,
 333                              filter_hash_bucket(priv, src_ip, dst_ip,
 334                                                 src_port, dst_port),
 335                              filter_chain) {
 336                 if (filter->src_ip == src_ip &&
 337                     filter->dst_ip == dst_ip &&
 338                     filter->ip_proto == ip_proto &&
 339                     filter->src_port == src_port &&
 340                     filter->dst_port == dst_port) {
 341                         ret = filter;
 342                         break;
 343                 }
 344         }
 345
 346         return ret;
 347 }
 348
 349 static int
 350 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
 351                    u16 rxq_index, u32 flow_id)
 352 {
 353         struct mlx4_en_priv *priv = netdev_priv(net_dev);
 354         struct mlx4_en_filter *filter;
 355         const struct iphdr *ip;
 356         const __be16 *ports;
 357         u8 ip_proto;
 358         __be32 src_ip;
 359         __be32 dst_ip;
 360         __be16 src_port;
 361         __be16 dst_port;
 362         int nhoff = skb_network_offset(skb);
 363         int ret = 0;
 364
 365         if (skb->protocol != htons(ETH_P_IP))
 366                 return -EPROTONOSUPPORT;
 367
 368         ip = (const struct iphdr *)(skb->data + nhoff);
 369         if (ip_is_fragment(ip))
 370                 return -EPROTONOSUPPORT;
 371
 372         if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
 373                 return -EPROTONOSUPPORT;
 374         ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
 375
 376         ip_proto = ip->protocol;
 377         src_ip = ip->saddr;
 378         dst_ip = ip->daddr;
 379         src_port = ports[0];
 380         dst_port = ports[1];
 381
 382         spin_lock_bh(&priv->filters_lock);
 383         filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
 384                                      src_port, dst_port);
 385         if (filter) {
 386                 if (filter->rxq_index == rxq_index)
 387                         goto out;
 388
 389                 filter->rxq_index = rxq_index;
 390         } else {
 391                 filter = mlx4_en_filter_alloc(priv, rxq_index,
 392                                               src_ip, dst_ip, ip_proto,
 393                                               src_port, dst_port, flow_id);
 394                 if (!filter) {
 395                         ret = -ENOMEM;
 396                         goto err;
 397                 }
 398         }
 399
 400         queue_work(priv->mdev->workqueue, &filter->work);
 401
 402 out:
 403         ret = filter->id;
 404 err:
 405         spin_unlock_bh(&priv->filters_lock);
 406
 407         return ret;
 408 }
 409
 410 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
 411 {
 412         struct mlx4_en_filter *filter, *tmp;
 413         LIST_HEAD(del_list);
 414
 415         spin_lock_bh(&priv->filters_lock);
 416         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 417                 list_move(&filter->next, &del_list);
 418                 hlist_del(&filter->filter_chain);
 419         }
 420         spin_unlock_bh(&priv->filters_lock);
 421
 422         list_for_each_entry_safe(filter, tmp, &del_list, next) {
 423                 cancel_work_sync(&filter->work);
 424                 mlx4_en_filter_free(filter);
 425         }
 426 }
 427
 428 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
 429 {
 430         struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
 431         LIST_HEAD(del_list);
 432         int i = 0;
 433
 434         spin_lock_bh(&priv->filters_lock);
 435         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 436                 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
 437                         break;
 438
 439                 if (filter->activated &&
 440                     !work_pending(&filter->work) &&
 441                     rps_may_expire_flow(priv->dev,
 442                                         filter->rxq_index, filter->flow_id,
 443                                         filter->id)) {
 444                         list_move(&filter->next, &del_list);
 445                         hlist_del(&filter->filter_chain);
 446                 } else
 447                         last_filter = filter;
 448
 449                 i++;
 450         }
 451
 452         if (last_filter && (&last_filter->next != priv->filters.next))
 453                 list_move(&priv->filters, &last_filter->next);
 454
 455         spin_unlock_bh(&priv->filters_lock);
 456
 457         list_for_each_entry_safe(filter, tmp, &del_list, next)
 458                 mlx4_en_filter_free(filter);
 459 }
 460 #endif
 461
 462 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
 463                                    __be16 proto, u16 vid)
 464 {
 465         struct mlx4_en_priv *priv = netdev_priv(dev);
 466         struct mlx4_en_dev *mdev = priv->mdev;
 467         int err;
 468         int idx;
 469
 470         en_dbg(HW, priv, "adding VLAN:%d\n", vid);
 471
 472         set_bit(vid, priv->active_vlans);
 473
 474         /* Add VID to port VLAN filter */
 475         mutex_lock(&mdev->state_lock);
 476         if (mdev->device_up && priv->port_up) {
 477                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 478                 if (err) {
 479                         en_err(priv, "Failed configuring VLAN filter\n");
 480                         goto out;
 481                 }
 482         }
 483         err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
 484         if (err)
 485                 en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
 486
 487 out:
 488         mutex_unlock(&mdev->state_lock);
 489         return err;
 490 }
 491
 492 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
 493                                     __be16 proto, u16 vid)
 494 {
 495         struct mlx4_en_priv *priv = netdev_priv(dev);
 496         struct mlx4_en_dev *mdev = priv->mdev;
 497         int err = 0;
 498
 499         en_dbg(HW, priv, "Killing VID:%d\n", vid);
 500
 501         clear_bit(vid, priv->active_vlans);
 502
 503         /* Remove VID from port VLAN filter */
 504         mutex_lock(&mdev->state_lock);
 505         mlx4_unregister_vlan(mdev->dev, priv->port, vid);
 506
 507         if (mdev->device_up && priv->port_up) {
 508                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 509                 if (err)
 510                         en_err(priv, "Failed configuring VLAN filter\n");
 511         }
 512         mutex_unlock(&mdev->state_lock);
 513
 514         return err;
 515 }
 516
 517 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
 518 {
 519         int i;
 520         for (i = ETH_ALEN - 1; i >= 0; --i) {
 521                 dst_mac[i] = src_mac & 0xff;
 522                 src_mac >>= 8;
 523         }
 524         memset(&dst_mac[ETH_ALEN], 0, 2);
 525 }
 526
 527
 528 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
 529                                     int qpn, u64 *reg_id)
 530 {
 531         int err;
 532
 533         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
 534             priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
 535                 return 0; /* do nothing */
 536
 537         err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
 538                                     MLX4_DOMAIN_NIC, reg_id);
 539         if (err) {
 540                 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
 541                 return err;
 542         }
 543         en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
 544         return 0;
 545 }
 546
 547
 548 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
 549                                 unsigned char *mac, int *qpn, u64 *reg_id)
 550 {
 551         struct mlx4_en_dev *mdev = priv->mdev;
 552         struct mlx4_dev *dev = mdev->dev;
 553         int err;
 554
 555         switch (dev->caps.steering_mode) {
 556         case MLX4_STEERING_MODE_B0: {
 557                 struct mlx4_qp qp;
 558                 u8 gid[16] = {0};
 559
 560                 qp.qpn = *qpn;
 561                 memcpy(&gid[10], mac, ETH_ALEN);
 562                 gid[5] = priv->port;
 563
 564                 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
 565                 break;
 566         }
 567         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 568                 struct mlx4_spec_list spec_eth = { {NULL} };
 569                 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 570
 571                 struct mlx4_net_trans_rule rule = {
 572                         .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 573                         .exclusive = 0,
 574                         .allow_loopback = 1,
 575                         .promisc_mode = MLX4_FS_REGULAR,
 576                         .priority = MLX4_DOMAIN_NIC,
 577                 };
 578
 579                 rule.port = priv->port;
 580                 rule.qpn = *qpn;
 581                 INIT_LIST_HEAD(&rule.list);
 582
 583                 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
 584                 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
 585                 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 586                 list_add_tail(&spec_eth.list, &rule.list);
 587
 588                 err = mlx4_flow_attach(dev, &rule, reg_id);
 589                 break;
 590         }
 591         default:
 592                 return -EINVAL;
 593         }
 594         if (err)
 595                 en_warn(priv, "Failed Attaching Unicast\n");
 596
 597         return err;
 598 }
 599
 600 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
 601                                      unsigned char *mac, int qpn, u64 reg_id)
 602 {
 603         struct mlx4_en_dev *mdev = priv->mdev;
 604         struct mlx4_dev *dev = mdev->dev;
 605
 606         switch (dev->caps.steering_mode) {
 607         case MLX4_STEERING_MODE_B0: {
 608                 struct mlx4_qp qp;
 609                 u8 gid[16] = {0};
 610
 611                 qp.qpn = qpn;
 612                 memcpy(&gid[10], mac, ETH_ALEN);
 613                 gid[5] = priv->port;
 614
 615                 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
 616                 break;
 617         }
 618         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 619                 mlx4_flow_detach(dev, reg_id);
 620                 break;
 621         }
 622         default:
 623                 en_err(priv, "Invalid steering mode.\n");
 624         }
 625 }
 626
 627 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
 628 {
 629         struct mlx4_en_dev *mdev = priv->mdev;
 630         struct mlx4_dev *dev = mdev->dev;
 631         int index = 0;
 632         int err = 0;
 633         int *qpn = &priv->base_qpn;
 634         u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 635
 636         en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
 637                priv->dev->dev_addr);
 638         index = mlx4_register_mac(dev, priv->port, mac);
 639         if (index < 0) {
 640                 err = index;
 641                 en_err(priv, "Failed adding MAC: %pM\n",
 642                        priv->dev->dev_addr);
 643                 return err;
 644         }
 645
 646         en_info(priv, "Steering Mode %d\n", dev->caps.steering_mode);
 647
 648         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 649                 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
 650                 *qpn = base_qpn + index;
 651                 return 0;
 652         }
 653
 654         err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP);
 655         en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
 656         if (err) {
 657                 en_err(priv, "Failed to reserve qp for mac registration\n");
 658                 mlx4_unregister_mac(dev, priv->port, mac);
 659                 return err;
 660         }
 661
 662         return 0;
 663 }
 664
 665 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
 666 {
 667         struct mlx4_en_dev *mdev = priv->mdev;
 668         struct mlx4_dev *dev = mdev->dev;
 669         int qpn = priv->base_qpn;
 670
 671         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 672                 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 673                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 674                        priv->dev->dev_addr);
 675                 mlx4_unregister_mac(dev, priv->port, mac);
 676         } else {
 677                 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
 678                        priv->port, qpn);
 679                 mlx4_qp_release_range(dev, qpn, 1);
 680                 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
 681         }
 682 }
 683
 684 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
 685                                unsigned char *new_mac, unsigned char *prev_mac)
 686 {
 687         struct mlx4_en_dev *mdev = priv->mdev;
 688         struct mlx4_dev *dev = mdev->dev;
 689         int err = 0;
 690         u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
 691
 692         if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
 693                 struct hlist_head *bucket;
 694                 unsigned int mac_hash;
 695                 struct mlx4_mac_entry *entry;
 696                 struct hlist_node *tmp;
 697                 u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
 698
 699                 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
 700                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 701                         if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
 702                                 mlx4_en_uc_steer_release(priv, entry->mac,
 703                                                          qpn, entry->reg_id);
 704                                 mlx4_unregister_mac(dev, priv->port,
 705                                                     prev_mac_u64);
 706                                 hlist_del_rcu(&entry->hlist);
 707                                 synchronize_rcu();
 708                                 memcpy(entry->mac, new_mac, ETH_ALEN);
 709                                 entry->reg_id = 0;
 710                                 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
 711                                 hlist_add_head_rcu(&entry->hlist,
 712                                                    &priv->mac_hash[mac_hash]);
 713                                 mlx4_register_mac(dev, priv->port, new_mac_u64);
 714                                 err = mlx4_en_uc_steer_add(priv, new_mac,
 715                                                            &qpn,
 716                                                            &entry->reg_id);
 717                                 if (err)
 718                                         return err;
 719                                 if (priv->tunnel_reg_id) {
 720                                         mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 721                                         priv->tunnel_reg_id = 0;
 722                                 }
 723                                 err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
 724                                                                &priv->tunnel_reg_id);
 725                                 return err;
 726                         }
 727                 }
 728                 return -EINVAL;
 729         }
 730
 731         return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
 732 }
 733
 734 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
 735                               unsigned char new_mac[ETH_ALEN + 2])
 736 {
 737         int err = 0;
 738
 739         if (priv->port_up) {
 740                 /* Remove old MAC and insert the new one */
 741                 err = mlx4_en_replace_mac(priv, priv->base_qpn,
 742                                           new_mac, priv->current_mac);
 743                 if (err)
 744                         en_err(priv, "Failed changing HW MAC address\n");
 745         } else
 746                 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
 747
 748         if (!err)
 749                 memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
 750
 751         return err;
 752 }
 753
 754 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
 755 {
 756         struct mlx4_en_priv *priv = netdev_priv(dev);
 757         struct mlx4_en_dev *mdev = priv->mdev;
 758         struct sockaddr *saddr = addr;
 759         unsigned char new_mac[ETH_ALEN + 2];
 760         int err;
 761
 762         if (!is_valid_ether_addr(saddr->sa_data))
 763                 return -EADDRNOTAVAIL;
 764
 765         mutex_lock(&mdev->state_lock);
 766         memcpy(new_mac, saddr->sa_data, ETH_ALEN);
 767         err = mlx4_en_do_set_mac(priv, new_mac);
 768         if (!err)
 769                 memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
 770         mutex_unlock(&mdev->state_lock);
 771
 772         return err;
 773 }
 774
 775 static void mlx4_en_clear_list(struct net_device *dev)
 776 {
 777         struct mlx4_en_priv *priv = netdev_priv(dev);
 778         struct mlx4_en_mc_list *tmp, *mc_to_del;
 779
 780         list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
 781                 list_del(&mc_to_del->list);
 782                 kfree(mc_to_del);
 783         }
 784 }
 785
 786 static void mlx4_en_cache_mclist(struct net_device *dev)
 787 {
 788         struct mlx4_en_priv *priv = netdev_priv(dev);
 789         struct netdev_hw_addr *ha;
 790         struct mlx4_en_mc_list *tmp;
 791
 792         mlx4_en_clear_list(dev);
 793         netdev_for_each_mc_addr(ha, dev) {
 794                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 795                 if (!tmp) {
 796                         mlx4_en_clear_list(dev);
 797                         return;
 798                 }
 799                 memcpy(tmp->addr, ha->addr, ETH_ALEN);
 800                 list_add_tail(&tmp->list, &priv->mc_list);
 801         }
 802 }
 803
 804 static void update_mclist_flags(struct mlx4_en_priv *priv,
 805                                 struct list_head *dst,
 806                                 struct list_head *src)
 807 {
 808         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 809         bool found;
 810
 811         /* Find all the entries that should be removed from dst,
 812          * These are the entries that are not found in src
 813          */
 814         list_for_each_entry(dst_tmp, dst, list) {
 815                 found = false;
 816                 list_for_each_entry(src_tmp, src, list) {
 817                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 818                                 found = true;
 819                                 break;
 820                         }
 821                 }
 822                 if (!found)
 823                         dst_tmp->action = MCLIST_REM;
 824         }
 825
 826         /* Add entries that exist in src but not in dst
 827          * mark them as need to add
 828          */
 829         list_for_each_entry(src_tmp, src, list) {
 830                 found = false;
 831                 list_for_each_entry(dst_tmp, dst, list) {
 832                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 833                                 dst_tmp->action = MCLIST_NONE;
 834                                 found = true;
 835                                 break;
 836                         }
 837                 }
 838                 if (!found) {
 839                         new_mc = kmemdup(src_tmp,
 840                                          sizeof(struct mlx4_en_mc_list),
 841                                          GFP_KERNEL);
 842                         if (!new_mc)
 843                                 return;
 844
 845                         new_mc->action = MCLIST_ADD;
 846                         list_add_tail(&new_mc->list, dst);
 847                 }
 848         }
 849 }
 850
 851 static void mlx4_en_set_rx_mode(struct net_device *dev)
 852 {
 853         struct mlx4_en_priv *priv = netdev_priv(dev);
 854
 855         if (!priv->port_up)
 856                 return;
 857
 858         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 859 }
 860
 861 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 862                                      struct mlx4_en_dev *mdev)
 863 {
 864         int err = 0;
 865
 866         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 867                 if (netif_msg_rx_status(priv))
 868                         en_warn(priv, "Entering promiscuous mode\n");
 869                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 870
 871                 /* Enable promiscouos mode */
 872                 switch (mdev->dev->caps.steering_mode) {
 873                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 874                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 875                                                           priv->port,
 876                                                           priv->base_qpn,
 877                                                           MLX4_FS_ALL_DEFAULT);
 878                         if (err)
 879                                 en_err(priv, "Failed enabling promiscuous mode\n");
 880                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 881                         break;
 882
 883                 case MLX4_STEERING_MODE_B0:
 884                         err = mlx4_unicast_promisc_add(mdev->dev,
 885                                                        priv->base_qpn,
 886                                                        priv->port);
 887                         if (err)
 888                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 889
 890                         /* Add the default qp number as multicast
 891                          * promisc
 892                          */
 893                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 894                                 err = mlx4_multicast_promisc_add(mdev->dev,
 895                                                                  priv->base_qpn,
 896                                                                  priv->port);
 897                                 if (err)
 898                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 899                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 900                         }
 901                         break;
 902
 903                 case MLX4_STEERING_MODE_A0:
 904                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 905                                                      priv->port,
 906                                                      priv->base_qpn,
 907                                                      1);
 908                         if (err)
 909                                 en_err(priv, "Failed enabling promiscuous mode\n");
 910                         break;
 911                 }
 912
 913                 /* Disable port multicast filter (unconditionally) */
 914                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 915                                           0, MLX4_MCAST_DISABLE);
 916                 if (err)
 917                         en_err(priv, "Failed disabling multicast filter\n");
 918         }
 919 }
 920
 921 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 922                                        struct mlx4_en_dev *mdev)
 923 {
 924         int err = 0;
 925
 926         if (netif_msg_rx_status(priv))
 927                 en_warn(priv, "Leaving promiscuous mode\n");
 928         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 929
 930         /* Disable promiscouos mode */
 931         switch (mdev->dev->caps.steering_mode) {
 932         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 933                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 934                                                      priv->port,
 935                                                      MLX4_FS_ALL_DEFAULT);
 936                 if (err)
 937                         en_err(priv, "Failed disabling promiscuous mode\n");
 938                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 939                 break;
 940
 941         case MLX4_STEERING_MODE_B0:
 942                 err = mlx4_unicast_promisc_remove(mdev->dev,
 943                                                   priv->base_qpn,
 944                                                   priv->port);
 945                 if (err)
 946                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 947                 /* Disable Multicast promisc */
 948                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 949                         err = mlx4_multicast_promisc_remove(mdev->dev,
 950                                                             priv->base_qpn,
 951                                                             priv->port);
 952                         if (err)
 953                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 954                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 955                 }
 956                 break;
 957
 958         case MLX4_STEERING_MODE_A0:
 959                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 960                                              priv->port,
 961                                              priv->base_qpn, 0);
 962                 if (err)
 963                         en_err(priv, "Failed disabling promiscuous mode\n");
 964                 break;
 965         }
 966 }
 967
 968 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
 969                                  struct net_device *dev,
 970                                  struct mlx4_en_dev *mdev)
 971 {
 972         struct mlx4_en_mc_list *mclist, *tmp;
 973         u64 mcast_addr = 0;
 974         u8 mc_list[16] = {0};
 975         int err = 0;
 976
 977         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
 978         if (dev->flags & IFF_ALLMULTI) {
 979                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 980                                           0, MLX4_MCAST_DISABLE);
 981                 if (err)
 982                         en_err(priv, "Failed disabling multicast filter\n");
 983
 984                 /* Add the default qp number as multicast promisc */
 985                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 986                         switch (mdev->dev->caps.steering_mode) {
 987                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 988                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
 989                                                                   priv->port,
 990                                                                   priv->base_qpn,
 991                                                                   MLX4_FS_MC_DEFAULT);
 992                                 break;
 993
 994                         case MLX4_STEERING_MODE_B0:
 995                                 err = mlx4_multicast_promisc_add(mdev->dev,
 996                                                                  priv->base_qpn,
 997                                                                  priv->port);
 998                                 break;
 999
1000                         case MLX4_STEERING_MODE_A0:
1001                                 break;
1002                         }
1003                         if (err)
1004                                 en_err(priv, "Failed entering multicast promisc mode\n");
1005                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
1006                 }
1007         } else {
1008                 /* Disable Multicast promisc */
1009                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1010                         switch (mdev->dev->caps.steering_mode) {
1011                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
1012                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
1013                                                                      priv->port,
1014                                                                      MLX4_FS_MC_DEFAULT);
1015                                 break;
1016
1017                         case MLX4_STEERING_MODE_B0:
1018                                 err = mlx4_multicast_promisc_remove(mdev->dev,
1019                                                                     priv->base_qpn,
1020                                                                     priv->port);
1021                                 break;
1022
1023                         case MLX4_STEERING_MODE_A0:
1024                                 break;
1025                         }
1026                         if (err)
1027                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
1028                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1029                 }
1030
1031                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1032                                           0, MLX4_MCAST_DISABLE);
1033                 if (err)
1034                         en_err(priv, "Failed disabling multicast filter\n");
1035
1036                 /* Flush mcast filter and init it with broadcast address */
1037                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
1038                                     1, MLX4_MCAST_CONFIG);
1039
1040                 /* Update multicast list - we cache all addresses so they won't
1041                  * change while HW is updated holding the command semaphor */
1042                 netif_addr_lock_bh(dev);
1043                 mlx4_en_cache_mclist(dev);
1044                 netif_addr_unlock_bh(dev);
1045                 list_for_each_entry(mclist, &priv->mc_list, list) {
1046                         mcast_addr = mlx4_mac_to_u64(mclist->addr);
1047                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
1048                                             mcast_addr, 0, MLX4_MCAST_CONFIG);
1049                 }
1050                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
1051                                           0, MLX4_MCAST_ENABLE);
1052                 if (err)
1053                         en_err(priv, "Failed enabling multicast filter\n");
1054
1055                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1056                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1057                         if (mclist->action == MCLIST_REM) {
1058                                 /* detach this address and delete from list */
1059                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1060                                 mc_list[5] = priv->port;
1061                                 err = mlx4_multicast_detach(mdev->dev,
1062                                                             priv->rss_map.indir_qp,
1063                                                             mc_list,
1064                                                             MLX4_PROT_ETH,
1065                                                             mclist->reg_id);
1066                                 if (err)
1067                                         en_err(priv, "Fail to detach multicast address\n");
1068
1069                                 if (mclist->tunnel_reg_id) {
1070                                         err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1071                                         if (err)
1072                                                 en_err(priv, "Failed to detach multicast address\n");
1073                                 }
1074
1075                                 /* remove from list */
1076                                 list_del(&mclist->list);
1077                                 kfree(mclist);
1078                         } else if (mclist->action == MCLIST_ADD) {
1079                                 /* attach the address */
1080                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1081                                 /* needed for B0 steering support */
1082                                 mc_list[5] = priv->port;
1083                                 err = mlx4_multicast_attach(mdev->dev,
1084                                                             priv->rss_map.indir_qp,
1085                                                             mc_list,
1086                                                             priv->port, 0,
1087                                                             MLX4_PROT_ETH,
1088                                                             &mclist->reg_id);
1089                                 if (err)
1090                                         en_err(priv, "Fail to attach multicast address\n");
1091
1092                                 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1093                                                                &mclist->tunnel_reg_id);
1094                                 if (err)
1095                                         en_err(priv, "Failed to attach multicast address\n");
1096                         }
1097                 }
1098         }
1099 }
1100
1101 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1102                                  struct net_device *dev,
1103                                  struct mlx4_en_dev *mdev)
1104 {
1105         struct netdev_hw_addr *ha;
1106         struct mlx4_mac_entry *entry;
1107         struct hlist_node *tmp;
1108         bool found;
1109         u64 mac;
1110         int err = 0;
1111         struct hlist_head *bucket;
1112         unsigned int i;
1113         int removed = 0;
1114         u32 prev_flags;
1115
1116         /* Note that we do not need to protect our mac_hash traversal with rcu,
1117          * since all modification code is protected by mdev->state_lock
1118          */
1119
1120         /* find what to remove */
1121         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1122                 bucket = &priv->mac_hash[i];
1123                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1124                         found = false;
1125                         netdev_for_each_uc_addr(ha, dev) {
1126                                 if (ether_addr_equal_64bits(entry->mac,
1127                                                             ha->addr)) {
1128                                         found = true;
1129                                         break;
1130                                 }
1131                         }
1132
1133                         /* MAC address of the port is not in uc list */
1134                         if (ether_addr_equal_64bits(entry->mac,
1135                                                     priv->current_mac))
1136                                 found = true;
1137
1138                         if (!found) {
1139                                 mac = mlx4_mac_to_u64(entry->mac);
1140                                 mlx4_en_uc_steer_release(priv, entry->mac,
1141                                                          priv->base_qpn,
1142                                                          entry->reg_id);
1143                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1144
1145                                 hlist_del_rcu(&entry->hlist);
1146                                 kfree_rcu(entry, rcu);
1147                                 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1148                                        entry->mac, priv->port);
1149                                 ++removed;
1150                         }
1151                 }
1152         }
1153
1154         /* if we didn't remove anything, there is no use in trying to add
1155          * again once we are in a forced promisc mode state
1156          */
1157         if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1158                 return;
1159
1160         prev_flags = priv->flags;
1161         priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1162
1163         /* find what to add */
1164         netdev_for_each_uc_addr(ha, dev) {
1165                 found = false;
1166                 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1167                 hlist_for_each_entry(entry, bucket, hlist) {
1168                         if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1169                                 found = true;
1170                                 break;
1171                         }
1172                 }
1173
1174                 if (!found) {
1175                         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1176                         if (!entry) {
1177                                 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1178                                        ha->addr, priv->port);
1179                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1180                                 break;
1181                         }
1182                         mac = mlx4_mac_to_u64(ha->addr);
1183                         memcpy(entry->mac, ha->addr, ETH_ALEN);
1184                         err = mlx4_register_mac(mdev->dev, priv->port, mac);
1185                         if (err < 0) {
1186                                 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1187                                        ha->addr, priv->port, err);
1188                                 kfree(entry);
1189                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1190                                 break;
1191                         }
1192                         err = mlx4_en_uc_steer_add(priv, ha->addr,
1193                                                    &priv->base_qpn,
1194                                                    &entry->reg_id);
1195                         if (err) {
1196                                 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1197                                        ha->addr, priv->port, err);
1198                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1199                                 kfree(entry);
1200                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1201                                 break;
1202                         } else {
1203                                 unsigned int mac_hash;
1204                                 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1205                                        ha->addr, priv->port);
1206                                 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1207                                 bucket = &priv->mac_hash[mac_hash];
1208                                 hlist_add_head_rcu(&entry->hlist, bucket);
1209                         }
1210                 }
1211         }
1212
1213         if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1214                 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1215                         priv->port);
1216         } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1217                 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1218                         priv->port);
1219         }
1220 }
1221
1222 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1223 {
1224         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1225                                                  rx_mode_task);
1226         struct mlx4_en_dev *mdev = priv->mdev;
1227         struct net_device *dev = priv->dev;
1228
1229         mutex_lock(&mdev->state_lock);
1230         if (!mdev->device_up) {
1231                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1232                 goto out;
1233         }
1234         if (!priv->port_up) {
1235                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1236                 goto out;
1237         }
1238
1239         if (!netif_carrier_ok(dev)) {
1240                 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1241                         if (priv->port_state.link_state) {
1242                                 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1243                                 netif_carrier_on(dev);
1244                                 en_dbg(LINK, priv, "Link Up\n");
1245                         }
1246                 }
1247         }
1248
1249         if (dev->priv_flags & IFF_UNICAST_FLT)
1250                 mlx4_en_do_uc_filter(priv, dev, mdev);
1251
1252         /* Promsicuous mode: disable all filters */
1253         if ((dev->flags & IFF_PROMISC) ||
1254             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1255                 mlx4_en_set_promisc_mode(priv, mdev);
1256                 goto out;
1257         }
1258
1259         /* Not in promiscuous mode */
1260         if (priv->flags & MLX4_EN_FLAG_PROMISC)
1261                 mlx4_en_clear_promisc_mode(priv, mdev);
1262
1263         mlx4_en_do_multicast(priv, dev, mdev);
1264 out:
1265         mutex_unlock(&mdev->state_lock);
1266 }
1267
1268 #ifdef CONFIG_NET_POLL_CONTROLLER
1269 static void mlx4_en_netpoll(struct net_device *dev)
1270 {
1271         struct mlx4_en_priv *priv = netdev_priv(dev);
1272         struct mlx4_en_cq *cq;
1273         int i;
1274
1275         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1276                 cq = priv->tx_cq[TX][i];
1277                 napi_schedule(&cq->napi);
1278         }
1279 }
1280 #endif
1281
1282 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1283 {
1284         u64 reg_id;
1285         int err = 0;
1286         int *qpn = &priv->base_qpn;
1287         struct mlx4_mac_entry *entry;
1288
1289         err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1290         if (err)
1291                 return err;
1292
1293         err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1294                                        &priv->tunnel_reg_id);
1295         if (err)
1296                 goto tunnel_err;
1297
1298         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1299         if (!entry) {
1300                 err = -ENOMEM;
1301                 goto alloc_err;
1302         }
1303
1304         memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1305         memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1306         entry->reg_id = reg_id;
1307         hlist_add_head_rcu(&entry->hlist,
1308                            &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1309
1310         return 0;
1311
1312 alloc_err:
1313         if (priv->tunnel_reg_id)
1314                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1315
1316 tunnel_err:
1317         mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1318         return err;
1319 }
1320
1321 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1322 {
1323         u64 mac;
1324         unsigned int i;
1325         int qpn = priv->base_qpn;
1326         struct hlist_head *bucket;
1327         struct hlist_node *tmp;
1328         struct mlx4_mac_entry *entry;
1329
1330         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1331                 bucket = &priv->mac_hash[i];
1332                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1333                         mac = mlx4_mac_to_u64(entry->mac);
1334                         en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1335                                entry->mac);
1336                         mlx4_en_uc_steer_release(priv, entry->mac,
1337                                                  qpn, entry->reg_id);
1338
1339                         mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1340                         hlist_del_rcu(&entry->hlist);
1341                         kfree_rcu(entry, rcu);
1342                 }
1343         }
1344
1345         if (priv->tunnel_reg_id) {
1346                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1347                 priv->tunnel_reg_id = 0;
1348         }
1349 }
1350
1351 static void mlx4_en_tx_timeout(struct net_device *dev)
1352 {
1353         struct mlx4_en_priv *priv = netdev_priv(dev);
1354         struct mlx4_en_dev *mdev = priv->mdev;
1355         int i;
1356
1357         if (netif_msg_timer(priv))
1358                 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1359
1360         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1361                 struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][i];
1362
1363                 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1364                         continue;
1365                 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1366                         i, tx_ring->qpn, tx_ring->sp_cqn,
1367                         tx_ring->cons, tx_ring->prod);
1368         }
1369
1370         priv->port_stats.tx_timeout++;
1371         en_dbg(DRV, priv, "Scheduling watchdog\n");
1372         queue_work(mdev->workqueue, &priv->watchdog_task);
1373 }
1374
1375
1376 static void
1377 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1378 {
1379         struct mlx4_en_priv *priv = netdev_priv(dev);
1380
1381         spin_lock_bh(&priv->stats_lock);
1382         mlx4_en_fold_software_stats(dev);
1383         netdev_stats_to_stats64(stats, &dev->stats);
1384         spin_unlock_bh(&priv->stats_lock);
1385 }
1386
1387 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1388 {
1389         struct mlx4_en_cq *cq;
1390         int i, t;
1391
1392         /* If we haven't received a specific coalescing setting
1393          * (module param), we set the moderation parameters as follows:
1394          * - moder_cnt is set to the number of mtu sized packets to
1395          *   satisfy our coalescing target.
1396          * - moder_time is set to a fixed value.
1397          */
1398         priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1399         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1400         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1401         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1402         en_dbg(INTR, priv, "Default coalescing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1403                priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1404
1405         /* Setup cq moderation params */
1406         for (i = 0; i < priv->rx_ring_num; i++) {
1407                 cq = priv->rx_cq[i];
1408                 cq->moder_cnt = priv->rx_frames;
1409                 cq->moder_time = priv->rx_usecs;
1410                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1411                 priv->last_moder_packets[i] = 0;
1412                 priv->last_moder_bytes[i] = 0;
1413         }
1414
1415         for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1416                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1417                         cq = priv->tx_cq[t][i];
1418                         cq->moder_cnt = priv->tx_frames;
1419                         cq->moder_time = priv->tx_usecs;
1420                 }
1421         }
1422
1423         /* Reset auto-moderation params */
1424         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1425         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1426         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1427         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1428         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1429         priv->adaptive_rx_coal = 1;
1430         priv->last_moder_jiffies = 0;
1431         priv->last_moder_tx_packets = 0;
1432 }
1433
1434 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1435 {
1436         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1437         u32 pkt_rate_high, pkt_rate_low;
1438         struct mlx4_en_cq *cq;
1439         unsigned long packets;
1440         unsigned long rate;
1441         unsigned long avg_pkt_size;
1442         unsigned long rx_packets;
1443         unsigned long rx_bytes;
1444         unsigned long rx_pkt_diff;
1445         int moder_time;
1446         int ring, err;
1447
1448         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1449                 return;
1450
1451         pkt_rate_low = READ_ONCE(priv->pkt_rate_low);
1452         pkt_rate_high = READ_ONCE(priv->pkt_rate_high);
1453
1454         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1455                 rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1456                 rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1457
1458                 rx_pkt_diff = rx_packets - priv->last_moder_packets[ring];
1459                 packets = rx_pkt_diff;
1460                 rate = packets * HZ / period;
1461                 avg_pkt_size = packets ? (rx_bytes -
1462                                 priv->last_moder_bytes[ring]) / packets : 0;
1463
1464                 /* Apply auto-moderation only when packet rate
1465                  * exceeds a rate that it matters */
1466                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1467                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1468                         if (rate <= pkt_rate_low)
1469                                 moder_time = priv->rx_usecs_low;
1470                         else if (rate >= pkt_rate_high)
1471                                 moder_time = priv->rx_usecs_high;
1472                         else
1473                                 moder_time = (rate - pkt_rate_low) *
1474                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1475                                         (pkt_rate_high - pkt_rate_low) +
1476                                         priv->rx_usecs_low;
1477                 } else {
1478                         moder_time = priv->rx_usecs_low;
1479                 }
1480
1481                 cq = priv->rx_cq[ring];
1482                 if (moder_time != priv->last_moder_time[ring] ||
1483                     cq->moder_cnt != priv->rx_frames) {
1484                         priv->last_moder_time[ring] = moder_time;
1485                         cq->moder_time = moder_time;
1486                         cq->moder_cnt = priv->rx_frames;
1487                         err = mlx4_en_set_cq_moder(priv, cq);
1488                         if (err)
1489                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1490                                        ring);
1491                 }
1492                 priv->last_moder_packets[ring] = rx_packets;
1493                 priv->last_moder_bytes[ring] = rx_bytes;
1494         }
1495
1496         priv->last_moder_jiffies = jiffies;
1497 }
1498
1499 static void mlx4_en_do_get_stats(struct work_struct *work)
1500 {
1501         struct delayed_work *delay = to_delayed_work(work);
1502         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1503                                                  stats_task);
1504         struct mlx4_en_dev *mdev = priv->mdev;
1505         int err;
1506
1507         mutex_lock(&mdev->state_lock);
1508         if (mdev->device_up) {
1509                 if (priv->port_up) {
1510                         err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1511                         if (err)
1512                                 en_dbg(HW, priv, "Could not update stats\n");
1513
1514                         mlx4_en_auto_moderation(priv);
1515                 }
1516
1517                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1518         }
1519         if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1520                 mlx4_en_do_set_mac(priv, priv->current_mac);
1521                 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1522         }
1523         mutex_unlock(&mdev->state_lock);
1524 }
1525
1526 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1527  * periodically
1528  */
1529 static void mlx4_en_service_task(struct work_struct *work)
1530 {
1531         struct delayed_work *delay = to_delayed_work(work);
1532         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1533                                                  service_task);
1534         struct mlx4_en_dev *mdev = priv->mdev;
1535
1536         mutex_lock(&mdev->state_lock);
1537         if (mdev->device_up) {
1538                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1539                         mlx4_en_ptp_overflow_check(mdev);
1540
1541                 mlx4_en_recover_from_oom(priv);
1542                 queue_delayed_work(mdev->workqueue, &priv->service_task,
1543                                    SERVICE_TASK_DELAY);
1544         }
1545         mutex_unlock(&mdev->state_lock);
1546 }
1547
1548 static void mlx4_en_linkstate(struct work_struct *work)
1549 {
1550         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1551                                                  linkstate_task);
1552         struct mlx4_en_dev *mdev = priv->mdev;
1553         int linkstate = priv->link_state;
1554
1555         mutex_lock(&mdev->state_lock);
1556         /* If observable port state changed set carrier state and
1557          * report to system log */
1558         if (priv->last_link_state != linkstate) {
1559                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1560                         en_info(priv, "Link Down\n");
1561                         netif_carrier_off(priv->dev);
1562                 } else {
1563                         en_info(priv, "Link Up\n");
1564                         netif_carrier_on(priv->dev);
1565                 }
1566         }
1567         priv->last_link_state = linkstate;
1568         mutex_unlock(&mdev->state_lock);
1569 }
1570
1571 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1572 {
1573         struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1574         int numa_node = priv->mdev->dev->numa_node;
1575
1576         if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1577                 return -ENOMEM;
1578
1579         cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1580                         ring->affinity_mask);
1581         return 0;
1582 }
1583
1584 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1585 {
1586         free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1587 }
1588
1589 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1590                                       int tx_ring_idx)
1591 {
1592         struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1593         int rr_index = tx_ring_idx;
1594
1595         tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1596         tx_ring->recycle_ring = priv->rx_ring[rr_index];
1597         en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1598                TX_XDP, tx_ring_idx, rr_index);
1599 }
1600
1601 int mlx4_en_start_port(struct net_device *dev)
1602 {
1603         struct mlx4_en_priv *priv = netdev_priv(dev);
1604         struct mlx4_en_dev *mdev = priv->mdev;
1605         struct mlx4_en_cq *cq;
1606         struct mlx4_en_tx_ring *tx_ring;
1607         int rx_index = 0;
1608         int err = 0;
1609         int i, t;
1610         int j;
1611         u8 mc_list[16] = {0};
1612
1613         if (priv->port_up) {
1614                 en_dbg(DRV, priv, "start port called while port already up\n");
1615                 return 0;
1616         }
1617
1618         INIT_LIST_HEAD(&priv->mc_list);
1619         INIT_LIST_HEAD(&priv->curr_list);
1620         INIT_LIST_HEAD(&priv->ethtool_list);
1621         memset(&priv->ethtool_rules[0], 0,
1622                sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1623
1624         /* Calculate Rx buf size */
1625         dev->mtu = min(dev->mtu, priv->max_mtu);
1626         mlx4_en_calc_rx_buf(dev);
1627         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1628
1629         /* Configure rx cq's and rings */
1630         err = mlx4_en_activate_rx_rings(priv);
1631         if (err) {
1632                 en_err(priv, "Failed to activate RX rings\n");
1633                 return err;
1634         }
1635         for (i = 0; i < priv->rx_ring_num; i++) {
1636                 cq = priv->rx_cq[i];
1637
1638                 err = mlx4_en_init_affinity_hint(priv, i);
1639                 if (err) {
1640                         en_err(priv, "Failed preparing IRQ affinity hint\n");
1641                         goto cq_err;
1642                 }
1643
1644                 err = mlx4_en_activate_cq(priv, cq, i);
1645                 if (err) {
1646                         en_err(priv, "Failed activating Rx CQ\n");
1647                         mlx4_en_free_affinity_hint(priv, i);
1648                         goto cq_err;
1649                 }
1650
1651                 for (j = 0; j < cq->size; j++) {
1652                         struct mlx4_cqe *cqe = NULL;
1653
1654                         cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1655                               priv->cqe_factor;
1656                         cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1657                 }
1658
1659                 err = mlx4_en_set_cq_moder(priv, cq);
1660                 if (err) {
1661                         en_err(priv, "Failed setting cq moderation parameters\n");
1662                         mlx4_en_deactivate_cq(priv, cq);
1663                         mlx4_en_free_affinity_hint(priv, i);
1664                         goto cq_err;
1665                 }
1666                 mlx4_en_arm_cq(priv, cq);
1667                 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1668                 ++rx_index;
1669         }
1670
1671         /* Set qp number */
1672         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1673         err = mlx4_en_get_qp(priv);
1674         if (err) {
1675                 en_err(priv, "Failed getting eth qp\n");
1676                 goto cq_err;
1677         }
1678         mdev->mac_removed[priv->port] = 0;
1679
1680         priv->counter_index =
1681                         mlx4_get_default_counter_index(mdev->dev, priv->port);
1682
1683         err = mlx4_en_config_rss_steer(priv);
1684         if (err) {
1685                 en_err(priv, "Failed configuring rss steering\n");
1686                 goto mac_err;
1687         }
1688
1689         err = mlx4_en_create_drop_qp(priv);
1690         if (err)
1691                 goto rss_err;
1692
1693         /* Configure tx cq's and rings */
1694         for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1695                 u8 num_tx_rings_p_up = t == TX ?
1696                         priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1697
1698                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1699                         /* Configure cq */
1700                         cq = priv->tx_cq[t][i];
1701                         err = mlx4_en_activate_cq(priv, cq, i);
1702                         if (err) {
1703                                 en_err(priv, "Failed allocating Tx CQ\n");
1704                                 goto tx_err;
1705                         }
1706                         err = mlx4_en_set_cq_moder(priv, cq);
1707                         if (err) {
1708                                 en_err(priv, "Failed setting cq moderation parameters\n");
1709                                 mlx4_en_deactivate_cq(priv, cq);
1710                                 goto tx_err;
1711                         }
1712                         en_dbg(DRV, priv,
1713                                "Resetting index of collapsed CQ:%d to -1\n", i);
1714                         cq->buf->wqe_index = cpu_to_be16(0xffff);
1715
1716                         /* Configure ring */
1717                         tx_ring = priv->tx_ring[t][i];
1718                         err = mlx4_en_activate_tx_ring(priv, tx_ring,
1719                                                        cq->mcq.cqn,
1720                                                        i / num_tx_rings_p_up);
1721                         if (err) {
1722                                 en_err(priv, "Failed allocating Tx ring\n");
1723                                 mlx4_en_deactivate_cq(priv, cq);
1724                                 goto tx_err;
1725                         }
1726                         if (t != TX_XDP) {
1727                                 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1728                                 tx_ring->recycle_ring = NULL;
1729
1730                                 /* Arm CQ for TX completions */
1731                                 mlx4_en_arm_cq(priv, cq);
1732
1733                         } else {
1734                                 mlx4_en_init_recycle_ring(priv, i);
1735                                 /* XDP TX CQ should never be armed */
1736                         }
1737
1738                         /* Set initial ownership of all Tx TXBBs to SW (1) */
1739                         for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1740                                 *((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1741                 }
1742         }
1743
1744         /* Configure port */
1745         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1746                                     priv->rx_skb_size + ETH_FCS_LEN,
1747                                     priv->prof->tx_pause,
1748                                     priv->prof->tx_ppp,
1749                                     priv->prof->rx_pause,
1750                                     priv->prof->rx_ppp);
1751         if (err) {
1752                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1753                        priv->port, err);
1754                 goto tx_err;
1755         }
1756
1757         err = mlx4_SET_PORT_user_mtu(mdev->dev, priv->port, dev->mtu);
1758         if (err) {
1759                 en_err(priv, "Failed to pass user MTU(%d) to Firmware for port %d, with error %d\n",
1760                        dev->mtu, priv->port, err);
1761                 goto tx_err;
1762         }
1763
1764         /* Set default qp number */
1765         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1766         if (err) {
1767                 en_err(priv, "Failed setting default qp numbers\n");
1768                 goto tx_err;
1769         }
1770
1771         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1772                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1773                 if (err) {
1774                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1775                                err);
1776                         goto tx_err;
1777                 }
1778         }
1779
1780         /* Init port */
1781         en_dbg(HW, priv, "Initializing port\n");
1782         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1783         if (err) {
1784                 en_err(priv, "Failed Initializing port\n");
1785                 goto tx_err;
1786         }
1787
1788         /* Set Unicast and VXLAN steering rules */
1789         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1790             mlx4_en_set_rss_steer_rules(priv))
1791                 mlx4_warn(mdev, "Failed setting steering rules\n");
1792
1793         /* Attach rx QP to bradcast address */
1794         eth_broadcast_addr(&mc_list[10]);
1795         mc_list[5] = priv->port; /* needed for B0 steering support */
1796         if (mlx4_multicast_attach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1797                                   priv->port, 0, MLX4_PROT_ETH,
1798                                   &priv->broadcast_id))
1799                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1800
1801         /* Must redo promiscuous mode setup. */
1802         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1803
1804         /* Schedule multicast task to populate multicast list */
1805         queue_work(mdev->workqueue, &priv->rx_mode_task);
1806
1807         if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1808                 udp_tunnel_get_rx_info(dev);
1809
1810         priv->port_up = true;
1811
1812         /* Process all completions if exist to prevent
1813          * the queues freezing if they are full
1814          */
1815         for (i = 0; i < priv->rx_ring_num; i++) {
1816                 local_bh_disable();
1817                 napi_schedule(&priv->rx_cq[i]->napi);
1818                 local_bh_enable();
1819         }
1820
1821         netif_tx_start_all_queues(dev);
1822         netif_device_attach(dev);
1823
1824         return 0;
1825
1826 tx_err:
1827         if (t == MLX4_EN_NUM_TX_TYPES) {
1828                 t--;
1829                 i = priv->tx_ring_num[t];
1830         }
1831         while (t >= 0) {
1832                 while (i--) {
1833                         mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1834                         mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1835                 }
1836                 if (!t--)
1837                         break;
1838                 i = priv->tx_ring_num[t];
1839         }
1840         mlx4_en_destroy_drop_qp(priv);
1841 rss_err:
1842         mlx4_en_release_rss_steer(priv);
1843 mac_err:
1844         mlx4_en_put_qp(priv);
1845 cq_err:
1846         while (rx_index--) {
1847                 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1848                 mlx4_en_free_affinity_hint(priv, rx_index);
1849         }
1850         for (i = 0; i < priv->rx_ring_num; i++)
1851                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1852
1853         return err; /* need to close devices */
1854 }
1855
1856
1857 void mlx4_en_stop_port(struct net_device *dev, int detach)
1858 {
1859         struct mlx4_en_priv *priv = netdev_priv(dev);
1860         struct mlx4_en_dev *mdev = priv->mdev;
1861         struct mlx4_en_mc_list *mclist, *tmp;
1862         struct ethtool_flow_id *flow, *tmp_flow;
1863         int i, t;
1864         u8 mc_list[16] = {0};
1865
1866         if (!priv->port_up) {
1867                 en_dbg(DRV, priv, "stop port called while port already down\n");
1868                 return;
1869         }
1870
1871         /* close port*/
1872         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1873
1874         /* Synchronize with tx routine */
1875         netif_tx_lock_bh(dev);
1876         if (detach)
1877                 netif_device_detach(dev);
1878         netif_tx_stop_all_queues(dev);
1879         netif_tx_unlock_bh(dev);
1880
1881         netif_tx_disable(dev);
1882
1883         spin_lock_bh(&priv->stats_lock);
1884         mlx4_en_fold_software_stats(dev);
1885         /* Set port as not active */
1886         priv->port_up = false;
1887         spin_unlock_bh(&priv->stats_lock);
1888
1889         priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1890
1891         /* Promsicuous mode */
1892         if (mdev->dev->caps.steering_mode ==
1893             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1894                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1895                                  MLX4_EN_FLAG_MC_PROMISC);
1896                 mlx4_flow_steer_promisc_remove(mdev->dev,
1897                                                priv->port,
1898                                                MLX4_FS_ALL_DEFAULT);
1899                 mlx4_flow_steer_promisc_remove(mdev->dev,
1900                                                priv->port,
1901                                                MLX4_FS_MC_DEFAULT);
1902         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1903                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1904
1905                 /* Disable promiscouos mode */
1906                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1907                                             priv->port);
1908
1909                 /* Disable Multicast promisc */
1910                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1911                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1912                                                       priv->port);
1913                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1914                 }
1915         }
1916
1917         /* Detach All multicasts */
1918         eth_broadcast_addr(&mc_list[10]);
1919         mc_list[5] = priv->port; /* needed for B0 steering support */
1920         mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp, mc_list,
1921                               MLX4_PROT_ETH, priv->broadcast_id);
1922         list_for_each_entry(mclist, &priv->curr_list, list) {
1923                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1924                 mc_list[5] = priv->port;
1925                 mlx4_multicast_detach(mdev->dev, priv->rss_map.indir_qp,
1926                                       mc_list, MLX4_PROT_ETH, mclist->reg_id);
1927                 if (mclist->tunnel_reg_id)
1928                         mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1929         }
1930         mlx4_en_clear_list(dev);
1931         list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1932                 list_del(&mclist->list);
1933                 kfree(mclist);
1934         }
1935
1936         /* Flush multicast filter */
1937         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1938
1939         /* Remove flow steering rules for the port*/
1940         if (mdev->dev->caps.steering_mode ==
1941             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1942                 ASSERT_RTNL();
1943                 list_for_each_entry_safe(flow, tmp_flow,
1944                                          &priv->ethtool_list, list) {
1945                         mlx4_flow_detach(mdev->dev, flow->id);
1946                         list_del(&flow->list);
1947                 }
1948         }
1949
1950         mlx4_en_destroy_drop_qp(priv);
1951
1952         /* Free TX Rings */
1953         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1954                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1955                         mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1956                         mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1957                 }
1958         }
1959         msleep(10);
1960
1961         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1962                 for (i = 0; i < priv->tx_ring_num[t]; i++)
1963                         mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1964
1965         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1966                 mlx4_en_delete_rss_steer_rules(priv);
1967
1968         /* Free RSS qps */
1969         mlx4_en_release_rss_steer(priv);
1970
1971         /* Unregister Mac address for the port */
1972         mlx4_en_put_qp(priv);
1973         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1974                 mdev->mac_removed[priv->port] = 1;
1975
1976         /* Free RX Rings */
1977         for (i = 0; i < priv->rx_ring_num; i++) {
1978                 struct mlx4_en_cq *cq = priv->rx_cq[i];
1979
1980                 napi_synchronize(&cq->napi);
1981                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1982                 mlx4_en_deactivate_cq(priv, cq);
1983
1984                 mlx4_en_free_affinity_hint(priv, i);
1985         }
1986 }
1987
1988 static void mlx4_en_restart(struct work_struct *work)
1989 {
1990         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1991                                                  watchdog_task);
1992         struct mlx4_en_dev *mdev = priv->mdev;
1993         struct net_device *dev = priv->dev;
1994
1995         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1996
1997         rtnl_lock();
1998         mutex_lock(&mdev->state_lock);
1999         if (priv->port_up) {
2000                 mlx4_en_stop_port(dev, 1);
2001                 if (mlx4_en_start_port(dev))
2002                         en_err(priv, "Failed restarting port %d\n", priv->port);
2003         }
2004         mutex_unlock(&mdev->state_lock);
2005         rtnl_unlock();
2006 }
2007
2008 static void mlx4_en_clear_stats(struct net_device *dev)
2009 {
2010         struct mlx4_en_priv *priv = netdev_priv(dev);
2011         struct mlx4_en_dev *mdev = priv->mdev;
2012         struct mlx4_en_tx_ring **tx_ring;
2013         int i;
2014
2015         if (!mlx4_is_slave(mdev->dev))
2016                 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
2017                         en_dbg(HW, priv, "Failed dumping statistics\n");
2018
2019         memset(&priv->pstats, 0, sizeof(priv->pstats));
2020         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
2021         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
2022         memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
2023         memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
2024         memset(&priv->rx_priority_flowstats, 0,
2025                sizeof(priv->rx_priority_flowstats));
2026         memset(&priv->tx_priority_flowstats, 0,
2027                sizeof(priv->tx_priority_flowstats));
2028         memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
2029
2030         tx_ring = priv->tx_ring[TX];
2031         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
2032                 tx_ring[i]->bytes = 0;
2033                 tx_ring[i]->packets = 0;
2034                 tx_ring[i]->tx_csum = 0;
2035                 tx_ring[i]->tx_dropped = 0;
2036                 tx_ring[i]->queue_stopped = 0;
2037                 tx_ring[i]->wake_queue = 0;
2038                 tx_ring[i]->tso_packets = 0;
2039                 tx_ring[i]->xmit_more = 0;
2040         }
2041         for (i = 0; i < priv->rx_ring_num; i++) {
2042                 priv->rx_ring[i]->bytes = 0;
2043                 priv->rx_ring[i]->packets = 0;
2044                 priv->rx_ring[i]->csum_ok = 0;
2045                 priv->rx_ring[i]->csum_none = 0;
2046                 priv->rx_ring[i]->csum_complete = 0;
2047         }
2048 }
2049
2050 static int mlx4_en_open(struct net_device *dev)
2051 {
2052         struct mlx4_en_priv *priv = netdev_priv(dev);
2053         struct mlx4_en_dev *mdev = priv->mdev;
2054         int err = 0;
2055
2056         mutex_lock(&mdev->state_lock);
2057
2058         if (!mdev->device_up) {
2059                 en_err(priv, "Cannot open - device down/disabled\n");
2060                 err = -EBUSY;
2061                 goto out;
2062         }
2063
2064         /* Reset HW statistics and SW counters */
2065         mlx4_en_clear_stats(dev);
2066
2067         err = mlx4_en_start_port(dev);
2068         if (err)
2069                 en_err(priv, "Failed starting port:%d\n", priv->port);
2070
2071 out:
2072         mutex_unlock(&mdev->state_lock);
2073         return err;
2074 }
2075
2076
2077 static int mlx4_en_close(struct net_device *dev)
2078 {
2079         struct mlx4_en_priv *priv = netdev_priv(dev);
2080         struct mlx4_en_dev *mdev = priv->mdev;
2081
2082         en_dbg(IFDOWN, priv, "Close port called\n");
2083
2084         mutex_lock(&mdev->state_lock);
2085
2086         mlx4_en_stop_port(dev, 0);
2087         netif_carrier_off(dev);
2088
2089         mutex_unlock(&mdev->state_lock);
2090         return 0;
2091 }
2092
2093 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2094 {
2095         int i, t;
2096
2097 #ifdef CONFIG_RFS_ACCEL
2098         priv->dev->rx_cpu_rmap = NULL;
2099 #endif
2100
2101         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2102                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2103                         if (priv->tx_ring[t] && priv->tx_ring[t][i])
2104                                 mlx4_en_destroy_tx_ring(priv,
2105                                                         &priv->tx_ring[t][i]);
2106                         if (priv->tx_cq[t] && priv->tx_cq[t][i])
2107                                 mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2108                 }
2109                 kfree(priv->tx_ring[t]);
2110                 kfree(priv->tx_cq[t]);
2111         }
2112
2113         for (i = 0; i < priv->rx_ring_num; i++) {
2114                 if (priv->rx_ring[i])
2115                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2116                                 priv->prof->rx_ring_size, priv->stride);
2117                 if (priv->rx_cq[i])
2118                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2119         }
2120
2121 }
2122
2123 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2124 {
2125         struct mlx4_en_port_profile *prof = priv->prof;
2126         int i, t;
2127         int node;
2128
2129         /* Create tx Rings */
2130         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2131                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2132                         node = cpu_to_node(i % num_online_cpus());
2133                         if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2134                                               prof->tx_ring_size, i, t, node))
2135                                 goto err;
2136
2137                         if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2138                                                    prof->tx_ring_size,
2139                                                    TXBB_SIZE, node, i))
2140                                 goto err;
2141                 }
2142         }
2143
2144         /* Create rx Rings */
2145         for (i = 0; i < priv->rx_ring_num; i++) {
2146                 node = cpu_to_node(i % num_online_cpus());
2147                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2148                                       prof->rx_ring_size, i, RX, node))
2149                         goto err;
2150
2151                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2152                                            prof->rx_ring_size, priv->stride,
2153                                            node))
2154                         goto err;
2155         }
2156
2157 #ifdef CONFIG_RFS_ACCEL
2158         priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2159 #endif
2160
2161         return 0;
2162
2163 err:
2164         en_err(priv, "Failed to allocate NIC resources\n");
2165         for (i = 0; i < priv->rx_ring_num; i++) {
2166                 if (priv->rx_ring[i])
2167                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2168                                                 prof->rx_ring_size,
2169                                                 priv->stride);
2170                 if (priv->rx_cq[i])
2171                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2172         }
2173         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2174                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2175                         if (priv->tx_ring[t][i])
2176                                 mlx4_en_destroy_tx_ring(priv,
2177                                                         &priv->tx_ring[t][i]);
2178                         if (priv->tx_cq[t][i])
2179                                 mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2180                 }
2181         }
2182         return -ENOMEM;
2183 }
2184
2185
2186 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2187                              struct mlx4_en_priv *src,
2188                              struct mlx4_en_port_profile *prof)
2189 {
2190         int t;
2191
2192         memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2193                sizeof(dst->hwtstamp_config));
2194         dst->num_tx_rings_p_up = prof->num_tx_rings_p_up;
2195         dst->rx_ring_num = prof->rx_ring_num;
2196         dst->flags = prof->flags;
2197         dst->mdev = src->mdev;
2198         dst->port = src->port;
2199         dst->dev = src->dev;
2200         dst->prof = prof;
2201         dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2202                                          DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2203
2204         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2205                 dst->tx_ring_num[t] = prof->tx_ring_num[t];
2206                 if (!dst->tx_ring_num[t])
2207                         continue;
2208
2209                 dst->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
2210                                           MAX_TX_RINGS, GFP_KERNEL);
2211                 if (!dst->tx_ring[t])
2212                         goto err_free_tx;
2213
2214                 dst->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
2215                                         MAX_TX_RINGS, GFP_KERNEL);
2216                 if (!dst->tx_cq[t]) {
2217                         kfree(dst->tx_ring[t]);
2218                         goto err_free_tx;
2219                 }
2220         }
2221
2222         return 0;
2223
2224 err_free_tx:
2225         while (t--) {
2226                 kfree(dst->tx_ring[t]);
2227                 kfree(dst->tx_cq[t]);
2228         }
2229         return -ENOMEM;
2230 }
2231
2232 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2233                                 struct mlx4_en_priv *src)
2234 {
2235         int t;
2236         memcpy(dst->rx_ring, src->rx_ring,
2237                sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2238         memcpy(dst->rx_cq, src->rx_cq,
2239                sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2240         memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2241                sizeof(dst->hwtstamp_config));
2242         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2243                 dst->tx_ring_num[t] = src->tx_ring_num[t];
2244                 dst->tx_ring[t] = src->tx_ring[t];
2245                 dst->tx_cq[t] = src->tx_cq[t];
2246         }
2247         dst->num_tx_rings_p_up = src->num_tx_rings_p_up;
2248         dst->rx_ring_num = src->rx_ring_num;
2249         memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2250 }
2251
2252 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2253                                 struct mlx4_en_priv *tmp,
2254                                 struct mlx4_en_port_profile *prof,
2255                                 bool carry_xdp_prog)
2256 {
2257         struct bpf_prog *xdp_prog;
2258         int i, t;
2259
2260         mlx4_en_copy_priv(tmp, priv, prof);
2261
2262         if (mlx4_en_alloc_resources(tmp)) {
2263                 en_warn(priv,
2264                         "%s: Resource allocation failed, using previous configuration\n",
2265                         __func__);
2266                 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2267                         kfree(tmp->tx_ring[t]);
2268                         kfree(tmp->tx_cq[t]);
2269                 }
2270                 return -ENOMEM;
2271         }
2272
2273         /* All rx_rings has the same xdp_prog.  Pick the first one. */
2274         xdp_prog = rcu_dereference_protected(
2275                 priv->rx_ring[0]->xdp_prog,
2276                 lockdep_is_held(&priv->mdev->state_lock));
2277
2278         if (xdp_prog && carry_xdp_prog) {
2279                 xdp_prog = bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2280                 if (IS_ERR(xdp_prog)) {
2281                         mlx4_en_free_resources(tmp);
2282                         return PTR_ERR(xdp_prog);
2283                 }
2284                 for (i = 0; i < tmp->rx_ring_num; i++)
2285                         rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2286                                            xdp_prog);
2287         }
2288
2289         return 0;
2290 }
2291
2292 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2293                                     struct mlx4_en_priv *tmp)
2294 {
2295         mlx4_en_free_resources(priv);
2296         mlx4_en_update_priv(priv, tmp);
2297 }
2298
2299 void mlx4_en_destroy_netdev(struct net_device *dev)
2300 {
2301         struct mlx4_en_priv *priv = netdev_priv(dev);
2302         struct mlx4_en_dev *mdev = priv->mdev;
2303
2304         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2305
2306         /* Unregister device - this will close the port if it was up */
2307         if (priv->registered) {
2308                 devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2309                                                               priv->port));
2310                 unregister_netdev(dev);
2311         }
2312
2313         if (priv->allocated)
2314                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2315
2316         cancel_delayed_work(&priv->stats_task);
2317         cancel_delayed_work(&priv->service_task);
2318         /* flush any pending task for this netdev */
2319         flush_workqueue(mdev->workqueue);
2320
2321         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2322                 mlx4_en_remove_timestamp(mdev);
2323
2324         /* Detach the netdev so tasks would not attempt to access it */
2325         mutex_lock(&mdev->state_lock);
2326         mdev->pndev[priv->port] = NULL;
2327         mdev->upper[priv->port] = NULL;
2328
2329 #ifdef CONFIG_RFS_ACCEL
2330         mlx4_en_cleanup_filters(priv);
2331 #endif
2332
2333         mlx4_en_free_resources(priv);
2334         mutex_unlock(&mdev->state_lock);
2335
2336         free_netdev(dev);
2337 }
2338
2339 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2340 {
2341         struct mlx4_en_priv *priv = netdev_priv(dev);
2342
2343         if (mtu > MLX4_EN_MAX_XDP_MTU) {
2344                 en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2345                        mtu, MLX4_EN_MAX_XDP_MTU);
2346                 return false;
2347         }
2348
2349         return true;
2350 }
2351
2352 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2353 {
2354         struct mlx4_en_priv *priv = netdev_priv(dev);
2355         struct mlx4_en_dev *mdev = priv->mdev;
2356         int err = 0;
2357
2358         en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2359                  dev->mtu, new_mtu);
2360
2361         if (priv->tx_ring_num[TX_XDP] &&
2362             !mlx4_en_check_xdp_mtu(dev, new_mtu))
2363                 return -EOPNOTSUPP;
2364
2365         dev->mtu = new_mtu;
2366
2367         if (netif_running(dev)) {
2368                 mutex_lock(&mdev->state_lock);
2369                 if (!mdev->device_up) {
2370                         /* NIC is probably restarting - let watchdog task reset
2371                          * the port */
2372                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2373                 } else {
2374                         mlx4_en_stop_port(dev, 1);
2375                         err = mlx4_en_start_port(dev);
2376                         if (err) {
2377                                 en_err(priv, "Failed restarting port:%d\n",
2378                                          priv->port);
2379                                 queue_work(mdev->workqueue, &priv->watchdog_task);
2380                         }
2381                 }
2382                 mutex_unlock(&mdev->state_lock);
2383         }
2384         return 0;
2385 }
2386
2387 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2388 {
2389         struct mlx4_en_priv *priv = netdev_priv(dev);
2390         struct mlx4_en_dev *mdev = priv->mdev;
2391         struct hwtstamp_config config;
2392
2393         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2394                 return -EFAULT;
2395
2396         /* reserved for future extensions */
2397         if (config.flags)
2398                 return -EINVAL;
2399
2400         /* device doesn't support time stamping */
2401         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2402                 return -EINVAL;
2403
2404         /* TX HW timestamp */
2405         switch (config.tx_type) {
2406         case HWTSTAMP_TX_OFF:
2407         case HWTSTAMP_TX_ON:
2408                 break;
2409         default:
2410                 return -ERANGE;
2411         }
2412
2413         /* RX HW timestamp */
2414         switch (config.rx_filter) {
2415         case HWTSTAMP_FILTER_NONE:
2416                 break;
2417         case HWTSTAMP_FILTER_ALL:
2418         case HWTSTAMP_FILTER_SOME:
2419         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2420         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2421         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2422         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2423         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2424         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2425         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2426         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2427         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2428         case HWTSTAMP_FILTER_PTP_V2_EVENT:
2429         case HWTSTAMP_FILTER_PTP_V2_SYNC:
2430         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2431         case HWTSTAMP_FILTER_NTP_ALL:
2432                 config.rx_filter = HWTSTAMP_FILTER_ALL;
2433                 break;
2434         default:
2435                 return -ERANGE;
2436         }
2437
2438         if (mlx4_en_reset_config(dev, config, dev->features)) {
2439                 config.tx_type = HWTSTAMP_TX_OFF;
2440                 config.rx_filter = HWTSTAMP_FILTER_NONE;
2441         }
2442
2443         return copy_to_user(ifr->ifr_data, &config,
2444                             sizeof(config)) ? -EFAULT : 0;
2445 }
2446
2447 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2448 {
2449         struct mlx4_en_priv *priv = netdev_priv(dev);
2450
2451         return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2452                             sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2453 }
2454
2455 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2456 {
2457         switch (cmd) {
2458         case SIOCSHWTSTAMP:
2459                 return mlx4_en_hwtstamp_set(dev, ifr);
2460         case SIOCGHWTSTAMP:
2461                 return mlx4_en_hwtstamp_get(dev, ifr);
2462         default:
2463                 return -EOPNOTSUPP;
2464         }
2465 }
2466
2467 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2468                                               netdev_features_t features)
2469 {
2470         struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2471         struct mlx4_en_dev *mdev = en_priv->mdev;
2472
2473         /* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2474          * enable/disable make sure S-TAG flag is always in same state as
2475          * C-TAG.
2476          */
2477         if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2478             !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2479                 features |= NETIF_F_HW_VLAN_STAG_RX;
2480         else
2481                 features &= ~NETIF_F_HW_VLAN_STAG_RX;
2482
2483         return features;
2484 }
2485
2486 static int mlx4_en_set_features(struct net_device *netdev,
2487                 netdev_features_t features)
2488 {
2489         struct mlx4_en_priv *priv = netdev_priv(netdev);
2490         bool reset = false;
2491         int ret = 0;
2492
2493         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2494                 en_info(priv, "Turn %s RX-FCS\n",
2495                         (features & NETIF_F_RXFCS) ? "ON" : "OFF");
2496                 reset = true;
2497         }
2498
2499         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2500                 u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2501
2502                 en_info(priv, "Turn %s RX-ALL\n",
2503                         ignore_fcs_value ? "ON" : "OFF");
2504                 ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2505                                               priv->port, ignore_fcs_value);
2506                 if (ret)
2507                         return ret;
2508         }
2509
2510         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2511                 en_info(priv, "Turn %s RX vlan strip offload\n",
2512                         (features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2513                 reset = true;
2514         }
2515
2516         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2517                 en_info(priv, "Turn %s TX vlan strip offload\n",
2518                         (features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2519
2520         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2521                 en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2522                         (features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2523
2524         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2525                 en_info(priv, "Turn %s loopback\n",
2526                         (features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2527                 mlx4_en_update_loopback_state(netdev, features);
2528         }
2529
2530         if (reset) {
2531                 ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2532                                            features);
2533                 if (ret)
2534                         return ret;
2535         }
2536
2537         return 0;
2538 }
2539
2540 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2541 {
2542         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2543         struct mlx4_en_dev *mdev = en_priv->mdev;
2544
2545         return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac);
2546 }
2547
2548 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2549                                __be16 vlan_proto)
2550 {
2551         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2552         struct mlx4_en_dev *mdev = en_priv->mdev;
2553
2554         return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2555                                 vlan_proto);
2556 }
2557
2558 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2559                                int max_tx_rate)
2560 {
2561         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2562         struct mlx4_en_dev *mdev = en_priv->mdev;
2563
2564         return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2565                                 max_tx_rate);
2566 }
2567
2568 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2569 {
2570         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2571         struct mlx4_en_dev *mdev = en_priv->mdev;
2572
2573         return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2574 }
2575
2576 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2577 {
2578         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2579         struct mlx4_en_dev *mdev = en_priv->mdev;
2580
2581         return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2582 }
2583
2584 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2585 {
2586         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2587         struct mlx4_en_dev *mdev = en_priv->mdev;
2588
2589         return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2590 }
2591
2592 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2593                                 struct ifla_vf_stats *vf_stats)
2594 {
2595         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2596         struct mlx4_en_dev *mdev = en_priv->mdev;
2597
2598         return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2599 }
2600
2601 #define PORT_ID_BYTE_LEN 8
2602 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2603                                     struct netdev_phys_item_id *ppid)
2604 {
2605         struct mlx4_en_priv *priv = netdev_priv(dev);
2606         struct mlx4_dev *mdev = priv->mdev->dev;
2607         int i;
2608         u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2609
2610         if (!phys_port_id)
2611                 return -EOPNOTSUPP;
2612
2613         ppid->id_len = sizeof(phys_port_id);
2614         for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2615                 ppid->id[i] =  phys_port_id & 0xff;
2616                 phys_port_id >>= 8;
2617         }
2618         return 0;
2619 }
2620
2621 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2622 {
2623         int ret;
2624         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2625                                                  vxlan_add_task);
2626
2627         ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2628         if (ret)
2629                 goto out;
2630
2631         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2632                                   VXLAN_STEER_BY_OUTER_MAC, 1);
2633 out:
2634         if (ret) {
2635                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2636                 return;
2637         }
2638
2639         /* set offloads */
2640         priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2641                                       NETIF_F_RXCSUM |
2642                                       NETIF_F_TSO | NETIF_F_TSO6 |
2643                                       NETIF_F_GSO_UDP_TUNNEL |
2644                                       NETIF_F_GSO_UDP_TUNNEL_CSUM |
2645                                       NETIF_F_GSO_PARTIAL;
2646 }
2647
2648 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2649 {
2650         int ret;
2651         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2652                                                  vxlan_del_task);
2653         /* unset offloads */
2654         priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2655                                         NETIF_F_RXCSUM |
2656                                         NETIF_F_TSO | NETIF_F_TSO6 |
2657                                         NETIF_F_GSO_UDP_TUNNEL |
2658                                         NETIF_F_GSO_UDP_TUNNEL_CSUM |
2659                                         NETIF_F_GSO_PARTIAL);
2660
2661         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2662                                   VXLAN_STEER_BY_OUTER_MAC, 0);
2663         if (ret)
2664                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2665
2666         priv->vxlan_port = 0;
2667 }
2668
2669 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2670                                    struct udp_tunnel_info *ti)
2671 {
2672         struct mlx4_en_priv *priv = netdev_priv(dev);
2673         __be16 port = ti->port;
2674         __be16 current_port;
2675
2676         if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2677                 return;
2678
2679         if (ti->sa_family != AF_INET)
2680                 return;
2681
2682         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2683                 return;
2684
2685         current_port = priv->vxlan_port;
2686         if (current_port && current_port != port) {
2687                 en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2688                         ntohs(current_port), ntohs(port));
2689                 return;
2690         }
2691
2692         priv->vxlan_port = port;
2693         queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2694 }
2695
2696 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2697                                    struct udp_tunnel_info *ti)
2698 {
2699         struct mlx4_en_priv *priv = netdev_priv(dev);
2700         __be16 port = ti->port;
2701         __be16 current_port;
2702
2703         if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2704                 return;
2705
2706         if (ti->sa_family != AF_INET)
2707                 return;
2708
2709         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2710                 return;
2711
2712         current_port = priv->vxlan_port;
2713         if (current_port != port) {
2714                 en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2715                 return;
2716         }
2717
2718         queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2719 }
2720
2721 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2722                                                 struct net_device *dev,
2723                                                 netdev_features_t features)
2724 {
2725         features = vlan_features_check(skb, features);
2726         features = vxlan_features_check(skb, features);
2727
2728         /* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2729          * support inner IPv6 checksums and segmentation so  we need to
2730          * strip that feature if this is an IPv6 encapsulated frame.
2731          */
2732         if (skb->encapsulation &&
2733             (skb->ip_summed == CHECKSUM_PARTIAL)) {
2734                 struct mlx4_en_priv *priv = netdev_priv(dev);
2735
2736                 if (!priv->vxlan_port ||
2737                     (ip_hdr(skb)->version != 4) ||
2738                     (udp_hdr(skb)->dest != priv->vxlan_port))
2739                         features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2740         }
2741
2742         return features;
2743 }
2744
2745 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2746 {
2747         struct mlx4_en_priv *priv = netdev_priv(dev);
2748         struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2749         struct mlx4_update_qp_params params;
2750         int err;
2751
2752         if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2753                 return -EOPNOTSUPP;
2754
2755         /* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2756         if (maxrate >> 12) {
2757                 params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2758                 params.rate_val  = maxrate / 1000;
2759         } else if (maxrate) {
2760                 params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2761                 params.rate_val  = maxrate;
2762         } else { /* zero serves to revoke the QP rate-limitation */
2763                 params.rate_unit = 0;
2764                 params.rate_val  = 0;
2765         }
2766
2767         err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2768                              &params);
2769         return err;
2770 }
2771
2772 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2773 {
2774         struct mlx4_en_priv *priv = netdev_priv(dev);
2775         struct mlx4_en_dev *mdev = priv->mdev;
2776         struct mlx4_en_port_profile new_prof;
2777         struct bpf_prog *old_prog;
2778         struct mlx4_en_priv *tmp;
2779         int tx_changed = 0;
2780         int xdp_ring_num;
2781         int port_up = 0;
2782         int err;
2783         int i;
2784
2785         xdp_ring_num = prog ? priv->rx_ring_num : 0;
2786
2787         /* No need to reconfigure buffers when simply swapping the
2788          * program for a new one.
2789          */
2790         if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2791                 if (prog) {
2792                         prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2793                         if (IS_ERR(prog))
2794                                 return PTR_ERR(prog);
2795                 }
2796                 mutex_lock(&mdev->state_lock);
2797                 for (i = 0; i < priv->rx_ring_num; i++) {
2798                         old_prog = rcu_dereference_protected(
2799                                         priv->rx_ring[i]->xdp_prog,
2800                                         lockdep_is_held(&mdev->state_lock));
2801                         rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2802                         if (old_prog)
2803                                 bpf_prog_put(old_prog);
2804                 }
2805                 mutex_unlock(&mdev->state_lock);
2806                 return 0;
2807         }
2808
2809         if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2810                 return -EOPNOTSUPP;
2811
2812         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2813         if (!tmp)
2814                 return -ENOMEM;
2815
2816         if (prog) {
2817                 prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2818                 if (IS_ERR(prog)) {
2819                         err = PTR_ERR(prog);
2820                         goto out;
2821                 }
2822         }
2823
2824         mutex_lock(&mdev->state_lock);
2825         memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2826         new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2827
2828         if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2829                 tx_changed = 1;
2830                 new_prof.tx_ring_num[TX] =
2831                         MAX_TX_RINGS - ALIGN(xdp_ring_num, priv->prof->num_up);
2832                 en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2833         }
2834
2835         err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2836         if (err) {
2837                 if (prog)
2838                         bpf_prog_sub(prog, priv->rx_ring_num - 1);
2839                 goto unlock_out;
2840         }
2841
2842         if (priv->port_up) {
2843                 port_up = 1;
2844                 mlx4_en_stop_port(dev, 1);
2845         }
2846
2847         mlx4_en_safe_replace_resources(priv, tmp);
2848         if (tx_changed)
2849                 netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2850
2851         for (i = 0; i < priv->rx_ring_num; i++) {
2852                 old_prog = rcu_dereference_protected(
2853                                         priv->rx_ring[i]->xdp_prog,
2854                                         lockdep_is_held(&mdev->state_lock));
2855                 rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2856                 if (old_prog)
2857                         bpf_prog_put(old_prog);
2858         }
2859
2860         if (port_up) {
2861                 err = mlx4_en_start_port(dev);
2862                 if (err) {
2863                         en_err(priv, "Failed starting port %d for XDP change\n",
2864                                priv->port);
2865                         queue_work(mdev->workqueue, &priv->watchdog_task);
2866                 }
2867         }
2868
2869 unlock_out:
2870         mutex_unlock(&mdev->state_lock);
2871 out:
2872         kfree(tmp);
2873         return err;
2874 }
2875
2876 static u32 mlx4_xdp_query(struct net_device *dev)
2877 {
2878         struct mlx4_en_priv *priv = netdev_priv(dev);
2879         struct mlx4_en_dev *mdev = priv->mdev;
2880         const struct bpf_prog *xdp_prog;
2881         u32 prog_id = 0;
2882
2883         if (!priv->tx_ring_num[TX_XDP])
2884                 return prog_id;
2885
2886         mutex_lock(&mdev->state_lock);
2887         xdp_prog = rcu_dereference_protected(
2888                 priv->rx_ring[0]->xdp_prog,
2889                 lockdep_is_held(&mdev->state_lock));
2890         if (xdp_prog)
2891                 prog_id = xdp_prog->aux->id;
2892         mutex_unlock(&mdev->state_lock);
2893
2894         return prog_id;
2895 }
2896
2897 static int mlx4_xdp(struct net_device *dev, struct netdev_xdp *xdp)
2898 {
2899         switch (xdp->command) {
2900         case XDP_SETUP_PROG:
2901                 return mlx4_xdp_set(dev, xdp->prog);
2902         case XDP_QUERY_PROG:
2903                 xdp->prog_id = mlx4_xdp_query(dev);
2904                 xdp->prog_attached = !!xdp->prog_id;
2905                 return 0;
2906         default:
2907                 return -EINVAL;
2908         }
2909 }
2910
2911 static const struct net_device_ops mlx4_netdev_ops = {
2912         .ndo_open               = mlx4_en_open,
2913         .ndo_stop               = mlx4_en_close,
2914         .ndo_start_xmit         = mlx4_en_xmit,
2915         .ndo_select_queue       = mlx4_en_select_queue,
2916         .ndo_get_stats64        = mlx4_en_get_stats64,
2917         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2918         .ndo_set_mac_address    = mlx4_en_set_mac,
2919         .ndo_validate_addr      = eth_validate_addr,
2920         .ndo_change_mtu         = mlx4_en_change_mtu,
2921         .ndo_do_ioctl           = mlx4_en_ioctl,
2922         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2923         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2924         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2925 #ifdef CONFIG_NET_POLL_CONTROLLER
2926         .ndo_poll_controller    = mlx4_en_netpoll,
2927 #endif
2928         .ndo_set_features       = mlx4_en_set_features,
2929         .ndo_fix_features       = mlx4_en_fix_features,
2930         .ndo_setup_tc           = __mlx4_en_setup_tc,
2931 #ifdef CONFIG_RFS_ACCEL
2932         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2933 #endif
2934         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2935         .ndo_udp_tunnel_add     = mlx4_en_add_vxlan_port,
2936         .ndo_udp_tunnel_del     = mlx4_en_del_vxlan_port,
2937         .ndo_features_check     = mlx4_en_features_check,
2938         .ndo_set_tx_maxrate     = mlx4_en_set_tx_maxrate,
2939         .ndo_xdp                = mlx4_xdp,
2940 };
2941
2942 static const struct net_device_ops mlx4_netdev_ops_master = {
2943         .ndo_open               = mlx4_en_open,
2944         .ndo_stop               = mlx4_en_close,
2945         .ndo_start_xmit         = mlx4_en_xmit,
2946         .ndo_select_queue       = mlx4_en_select_queue,
2947         .ndo_get_stats64        = mlx4_en_get_stats64,
2948         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2949         .ndo_set_mac_address    = mlx4_en_set_mac,
2950         .ndo_validate_addr      = eth_validate_addr,
2951         .ndo_change_mtu         = mlx4_en_change_mtu,
2952         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2953         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2954         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2955         .ndo_set_vf_mac         = mlx4_en_set_vf_mac,
2956         .ndo_set_vf_vlan        = mlx4_en_set_vf_vlan,
2957         .ndo_set_vf_rate        = mlx4_en_set_vf_rate,
2958         .ndo_set_vf_spoofchk    = mlx4_en_set_vf_spoofchk,
2959         .ndo_set_vf_link_state  = mlx4_en_set_vf_link_state,
2960         .ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2961         .ndo_get_vf_config      = mlx4_en_get_vf_config,
2962 #ifdef CONFIG_NET_POLL_CONTROLLER
2963         .ndo_poll_controller    = mlx4_en_netpoll,
2964 #endif
2965         .ndo_set_features       = mlx4_en_set_features,
2966         .ndo_fix_features       = mlx4_en_fix_features,
2967         .ndo_setup_tc           = __mlx4_en_setup_tc,
2968 #ifdef CONFIG_RFS_ACCEL
2969         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2970 #endif
2971         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2972         .ndo_udp_tunnel_add     = mlx4_en_add_vxlan_port,
2973         .ndo_udp_tunnel_del     = mlx4_en_del_vxlan_port,
2974         .ndo_features_check     = mlx4_en_features_check,
2975         .ndo_set_tx_maxrate     = mlx4_en_set_tx_maxrate,
2976         .ndo_xdp                = mlx4_xdp,
2977 };
2978
2979 struct mlx4_en_bond {
2980         struct work_struct work;
2981         struct mlx4_en_priv *priv;
2982         int is_bonded;
2983         struct mlx4_port_map port_map;
2984 };
2985
2986 static void mlx4_en_bond_work(struct work_struct *work)
2987 {
2988         struct mlx4_en_bond *bond = container_of(work,
2989                                                      struct mlx4_en_bond,
2990                                                      work);
2991         int err = 0;
2992         struct mlx4_dev *dev = bond->priv->mdev->dev;
2993
2994         if (bond->is_bonded) {
2995                 if (!mlx4_is_bonded(dev)) {
2996                         err = mlx4_bond(dev);
2997                         if (err)
2998                                 en_err(bond->priv, "Fail to bond device\n");
2999                 }
3000                 if (!err) {
3001                         err = mlx4_port_map_set(dev, &bond->port_map);
3002                         if (err)
3003                                 en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
3004                                        bond->port_map.port1,
3005                                        bond->port_map.port2,
3006                                        err);
3007                 }
3008         } else if (mlx4_is_bonded(dev)) {
3009                 err = mlx4_unbond(dev);
3010                 if (err)
3011                         en_err(bond->priv, "Fail to unbond device\n");
3012         }
3013         dev_put(bond->priv->dev);
3014         kfree(bond);
3015 }
3016
3017 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
3018                                    u8 v2p_p1, u8 v2p_p2)
3019 {
3020         struct mlx4_en_bond *bond = NULL;
3021
3022         bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
3023         if (!bond)
3024                 return -ENOMEM;
3025
3026         INIT_WORK(&bond->work, mlx4_en_bond_work);
3027         bond->priv = priv;
3028         bond->is_bonded = is_bonded;
3029         bond->port_map.port1 = v2p_p1;
3030         bond->port_map.port2 = v2p_p2;
3031         dev_hold(priv->dev);
3032         queue_work(priv->mdev->workqueue, &bond->work);
3033         return 0;
3034 }
3035
3036 int mlx4_en_netdev_event(struct notifier_block *this,
3037                          unsigned long event, void *ptr)
3038 {
3039         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
3040         u8 port = 0;
3041         struct mlx4_en_dev *mdev;
3042         struct mlx4_dev *dev;
3043         int i, num_eth_ports = 0;
3044         bool do_bond = true;
3045         struct mlx4_en_priv *priv;
3046         u8 v2p_port1 = 0;
3047         u8 v2p_port2 = 0;
3048
3049         if (!net_eq(dev_net(ndev), &init_net))
3050                 return NOTIFY_DONE;
3051
3052         mdev = container_of(this, struct mlx4_en_dev, nb);
3053         dev = mdev->dev;
3054
3055         /* Go into this mode only when two network devices set on two ports
3056          * of the same mlx4 device are slaves of the same bonding master
3057          */
3058         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
3059                 ++num_eth_ports;
3060                 if (!port && (mdev->pndev[i] == ndev))
3061                         port = i;
3062                 mdev->upper[i] = mdev->pndev[i] ?
3063                         netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
3064                 /* condition not met: network device is a slave */
3065                 if (!mdev->upper[i])
3066                         do_bond = false;
3067                 if (num_eth_ports < 2)
3068                         continue;
3069                 /* condition not met: same master */
3070                 if (mdev->upper[i] != mdev->upper[i-1])
3071                         do_bond = false;
3072         }
3073         /* condition not met: 2 salves */
3074         do_bond = (num_eth_ports ==  2) ? do_bond : false;
3075
3076         /* handle only events that come with enough info */
3077         if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
3078                 return NOTIFY_DONE;
3079
3080         priv = netdev_priv(ndev);
3081         if (do_bond) {
3082                 struct netdev_notifier_bonding_info *notifier_info = ptr;
3083                 struct netdev_bonding_info *bonding_info =
3084                         &notifier_info->bonding_info;
3085
3086                 /* required mode 1, 2 or 4 */
3087                 if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3088                     (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
3089                     (bonding_info->master.bond_mode != BOND_MODE_8023AD))
3090                         do_bond = false;
3091
3092                 /* require exactly 2 slaves */
3093                 if (bonding_info->master.num_slaves != 2)
3094                         do_bond = false;
3095
3096                 /* calc v2p */
3097                 if (do_bond) {
3098                         if (bonding_info->master.bond_mode ==
3099                             BOND_MODE_ACTIVEBACKUP) {
3100                                 /* in active-backup mode virtual ports are
3101                                  * mapped to the physical port of the active
3102                                  * slave */
3103                                 if (bonding_info->slave.state ==
3104                                     BOND_STATE_BACKUP) {
3105                                         if (port == 1) {
3106                                                 v2p_port1 = 2;
3107                                                 v2p_port2 = 2;
3108                                         } else {
3109                                                 v2p_port1 = 1;
3110                                                 v2p_port2 = 1;
3111                                         }
3112                                 } else { /* BOND_STATE_ACTIVE */
3113                                         if (port == 1) {
3114                                                 v2p_port1 = 1;
3115                                                 v2p_port2 = 1;
3116                                         } else {
3117                                                 v2p_port1 = 2;
3118                                                 v2p_port2 = 2;
3119                                         }
3120                                 }
3121                         } else { /* Active-Active */
3122                                 /* in active-active mode a virtual port is
3123                                  * mapped to the native physical port if and only
3124                                  * if the physical port is up */
3125                                 __s8 link = bonding_info->slave.link;
3126
3127                                 if (port == 1)
3128                                         v2p_port2 = 2;
3129                                 else
3130                                         v2p_port1 = 1;
3131                                 if ((link == BOND_LINK_UP) ||
3132                                     (link == BOND_LINK_FAIL)) {
3133                                         if (port == 1)
3134                                                 v2p_port1 = 1;
3135                                         else
3136                                                 v2p_port2 = 2;
3137                                 } else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3138                                         if (port == 1)
3139                                                 v2p_port1 = 2;
3140                                         else
3141                                                 v2p_port2 = 1;
3142                                 }
3143                         }
3144                 }
3145         }
3146
3147         mlx4_en_queue_bond_work(priv, do_bond,
3148                                 v2p_port1, v2p_port2);
3149
3150         return NOTIFY_DONE;
3151 }
3152
3153 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3154                                      struct mlx4_en_stats_bitmap *stats_bitmap,
3155                                      u8 rx_ppp, u8 rx_pause,
3156                                      u8 tx_ppp, u8 tx_pause)
3157 {
3158         int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3159
3160         if (!mlx4_is_slave(dev) &&
3161             (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3162                 mutex_lock(&stats_bitmap->mutex);
3163                 bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3164
3165                 if (rx_ppp)
3166                         bitmap_set(stats_bitmap->bitmap, last_i,
3167                                    NUM_FLOW_PRIORITY_STATS_RX);
3168                 last_i += NUM_FLOW_PRIORITY_STATS_RX;
3169
3170                 if (rx_pause && !(rx_ppp))
3171                         bitmap_set(stats_bitmap->bitmap, last_i,
3172                                    NUM_FLOW_STATS_RX);
3173                 last_i += NUM_FLOW_STATS_RX;
3174
3175                 if (tx_ppp)
3176                         bitmap_set(stats_bitmap->bitmap, last_i,
3177                                    NUM_FLOW_PRIORITY_STATS_TX);
3178                 last_i += NUM_FLOW_PRIORITY_STATS_TX;
3179
3180                 if (tx_pause && !(tx_ppp))
3181                         bitmap_set(stats_bitmap->bitmap, last_i,
3182                                    NUM_FLOW_STATS_TX);
3183                 last_i += NUM_FLOW_STATS_TX;
3184
3185                 mutex_unlock(&stats_bitmap->mutex);
3186         }
3187 }
3188
3189 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3190                               struct mlx4_en_stats_bitmap *stats_bitmap,
3191                               u8 rx_ppp, u8 rx_pause,
3192                               u8 tx_ppp, u8 tx_pause)
3193 {
3194         int last_i = 0;
3195
3196         mutex_init(&stats_bitmap->mutex);
3197         bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3198
3199         if (mlx4_is_slave(dev)) {
3200                 bitmap_set(stats_bitmap->bitmap, last_i +
3201                                          MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3202                 bitmap_set(stats_bitmap->bitmap, last_i +
3203                                          MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3204                 bitmap_set(stats_bitmap->bitmap, last_i +
3205                                          MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3206                 bitmap_set(stats_bitmap->bitmap, last_i +
3207                                          MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3208                 bitmap_set(stats_bitmap->bitmap, last_i +
3209                                          MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3210                 bitmap_set(stats_bitmap->bitmap, last_i +
3211                                          MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3212         } else {
3213                 bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3214         }
3215         last_i += NUM_MAIN_STATS;
3216
3217         bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3218         last_i += NUM_PORT_STATS;
3219
3220         if (mlx4_is_master(dev))
3221                 bitmap_set(stats_bitmap->bitmap, last_i,
3222                            NUM_PF_STATS);
3223         last_i += NUM_PF_STATS;
3224
3225         mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3226                                         rx_ppp, rx_pause,
3227                                         tx_ppp, tx_pause);
3228         last_i += NUM_FLOW_STATS;
3229
3230         if (!mlx4_is_slave(dev))
3231                 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3232         last_i += NUM_PKT_STATS;
3233
3234         bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3235         last_i += NUM_XDP_STATS;
3236 }
3237
3238 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3239                         struct mlx4_en_port_profile *prof)
3240 {
3241         struct net_device *dev;
3242         struct mlx4_en_priv *priv;
3243         int i, t;
3244         int err;
3245
3246         dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3247                                  MAX_TX_RINGS, MAX_RX_RINGS);
3248         if (dev == NULL)
3249                 return -ENOMEM;
3250
3251         netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3252         netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3253
3254         SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3255         dev->dev_port = port - 1;
3256
3257         /*
3258          * Initialize driver private data
3259          */
3260
3261         priv = netdev_priv(dev);
3262         memset(priv, 0, sizeof(struct mlx4_en_priv));
3263         priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3264         spin_lock_init(&priv->stats_lock);
3265         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3266         INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
3267         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3268         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3269         INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3270         INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
3271         INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
3272 #ifdef CONFIG_RFS_ACCEL
3273         INIT_LIST_HEAD(&priv->filters);
3274         spin_lock_init(&priv->filters_lock);
3275 #endif
3276
3277         priv->dev = dev;
3278         priv->mdev = mdev;
3279         priv->ddev = &mdev->pdev->dev;
3280         priv->prof = prof;
3281         priv->port = port;
3282         priv->port_up = false;
3283         priv->flags = prof->flags;
3284         priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3285         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3286                         MLX4_WQE_CTRL_SOLICITED);
3287         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
3288         priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3289         netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3290
3291         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3292                 priv->tx_ring_num[t] = prof->tx_ring_num[t];
3293                 if (!priv->tx_ring_num[t])
3294                         continue;
3295
3296                 priv->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
3297                                            MAX_TX_RINGS, GFP_KERNEL);
3298                 if (!priv->tx_ring[t]) {
3299                         err = -ENOMEM;
3300                         goto err_free_tx;
3301                 }
3302                 priv->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
3303                                          MAX_TX_RINGS, GFP_KERNEL);
3304                 if (!priv->tx_cq[t]) {
3305                         kfree(priv->tx_ring[t]);
3306                         err = -ENOMEM;
3307                         goto out;
3308                 }
3309         }
3310         priv->rx_ring_num = prof->rx_ring_num;
3311         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3312         priv->cqe_size = mdev->dev->caps.cqe_size;
3313         priv->mac_index = -1;
3314         priv->msg_enable = MLX4_EN_MSG_LEVEL;
3315 #ifdef CONFIG_MLX4_EN_DCB
3316         if (!mlx4_is_slave(priv->mdev->dev)) {
3317                 priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3318                         DCB_CAP_DCBX_VER_IEEE;
3319                 priv->flags |= MLX4_EN_DCB_ENABLED;
3320                 priv->cee_config.pfc_state = false;
3321
3322                 for (i = 0; i < MLX4_EN_NUM_UP_HIGH; i++)
3323                         priv->cee_config.dcb_pfc[i] = pfc_disabled;
3324
3325                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3326                         dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3327                 } else {
3328                         en_info(priv, "enabling only PFC DCB ops\n");
3329                         dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3330                 }
3331         }
3332 #endif
3333
3334         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3335                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
3336
3337         /* Query for default mac and max mtu */
3338         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3339
3340         if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3341             MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3342                 priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3343
3344         /* Set default MAC */
3345         dev->addr_len = ETH_ALEN;
3346         mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3347         if (!is_valid_ether_addr(dev->dev_addr)) {
3348                 en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
3349                        priv->port, dev->dev_addr);
3350                 err = -EINVAL;
3351                 goto out;
3352         } else if (mlx4_is_slave(priv->mdev->dev) &&
3353                    (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3354                 /* Random MAC was assigned in mlx4_slave_cap
3355                  * in mlx4_core module
3356                  */
3357                 dev->addr_assign_type |= NET_ADDR_RANDOM;
3358                 en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3359         }
3360
3361         memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3362
3363         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3364                                           DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3365         err = mlx4_en_alloc_resources(priv);
3366         if (err)
3367                 goto out;
3368
3369         /* Initialize time stamping config */
3370         priv->hwtstamp_config.flags = 0;
3371         priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3372         priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3373
3374         /* Allocate page for receive rings */
3375         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3376                                 MLX4_EN_PAGE_SIZE);
3377         if (err) {
3378                 en_err(priv, "Failed to allocate page for rx qps\n");
3379                 goto out;
3380         }
3381         priv->allocated = 1;
3382
3383         /*
3384          * Initialize netdev entry points
3385          */
3386         if (mlx4_is_master(priv->mdev->dev))
3387                 dev->netdev_ops = &mlx4_netdev_ops_master;
3388         else
3389                 dev->netdev_ops = &mlx4_netdev_ops;
3390         dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3391         netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3392         netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3393
3394         dev->ethtool_ops = &mlx4_en_ethtool_ops;
3395
3396         /*
3397          * Set driver features
3398          */
3399         dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3400         if (mdev->LSO_support)
3401                 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3402
3403         dev->vlan_features = dev->hw_features;
3404
3405         dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3406         dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3407                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3408                         NETIF_F_HW_VLAN_CTAG_FILTER;
3409         dev->hw_features |= NETIF_F_LOOPBACK |
3410                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3411
3412         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3413                 dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3414                         NETIF_F_HW_VLAN_STAG_FILTER;
3415                 dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3416         }
3417
3418         if (mlx4_is_slave(mdev->dev)) {
3419                 bool vlan_offload_disabled;
3420                 int phv;
3421
3422                 err = get_phv_bit(mdev->dev, port, &phv);
3423                 if (!err && phv) {
3424                         dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3425                         priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3426                 }
3427                 err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3428                                                         &vlan_offload_disabled);
3429                 if (!err && vlan_offload_disabled) {
3430                         dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3431                                               NETIF_F_HW_VLAN_CTAG_RX |
3432                                               NETIF_F_HW_VLAN_STAG_TX |
3433                                               NETIF_F_HW_VLAN_STAG_RX);
3434                         dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3435                                            NETIF_F_HW_VLAN_CTAG_RX |
3436                                            NETIF_F_HW_VLAN_STAG_TX |
3437                                            NETIF_F_HW_VLAN_STAG_RX);
3438                 }
3439         } else {
3440                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3441                     !(mdev->dev->caps.flags2 &
3442                       MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3443                         dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3444         }
3445
3446         if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3447                 dev->hw_features |= NETIF_F_RXFCS;
3448
3449         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3450                 dev->hw_features |= NETIF_F_RXALL;
3451
3452         if (mdev->dev->caps.steering_mode ==
3453             MLX4_STEERING_MODE_DEVICE_MANAGED &&
3454             mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3455                 dev->hw_features |= NETIF_F_NTUPLE;
3456
3457         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3458                 dev->priv_flags |= IFF_UNICAST_FLT;
3459
3460         /* Setting a default hash function value */
3461         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3462                 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3463         } else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3464                 priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3465         } else {
3466                 en_warn(priv,
3467                         "No RSS hash capabilities exposed, using Toeplitz\n");
3468                 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3469         }
3470
3471         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3472                 dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3473                                     NETIF_F_GSO_UDP_TUNNEL_CSUM |
3474                                     NETIF_F_GSO_PARTIAL;
3475                 dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3476                                     NETIF_F_GSO_UDP_TUNNEL_CSUM |
3477                                     NETIF_F_GSO_PARTIAL;
3478                 dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3479         }
3480
3481         /* MTU range: 46 - hw-specific max */
3482         dev->min_mtu = MLX4_EN_MIN_MTU;
3483         dev->max_mtu = priv->max_mtu;
3484
3485         mdev->pndev[port] = dev;
3486         mdev->upper[port] = NULL;
3487
3488         netif_carrier_off(dev);
3489         mlx4_en_set_default_moderation(priv);
3490
3491         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3492         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3493
3494         mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3495
3496         /* Configure port */
3497         mlx4_en_calc_rx_buf(dev);
3498         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3499                                     priv->rx_skb_size + ETH_FCS_LEN,
3500                                     prof->tx_pause, prof->tx_ppp,
3501                                     prof->rx_pause, prof->rx_ppp);
3502         if (err) {
3503                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3504                        priv->port, err);
3505                 goto out;
3506         }
3507
3508         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3509                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3510                 if (err) {
3511                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3512                                err);
3513                         goto out;
3514                 }
3515         }
3516
3517         /* Init port */
3518         en_warn(priv, "Initializing port\n");
3519         err = mlx4_INIT_PORT(mdev->dev, priv->port);
3520         if (err) {
3521                 en_err(priv, "Failed Initializing port\n");
3522                 goto out;
3523         }
3524         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3525
3526         /* Initialize time stamp mechanism */
3527         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3528                 mlx4_en_init_timestamp(mdev);
3529
3530         queue_delayed_work(mdev->workqueue, &priv->service_task,
3531                            SERVICE_TASK_DELAY);
3532
3533         mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3534                                  mdev->profile.prof[priv->port].rx_ppp,
3535                                  mdev->profile.prof[priv->port].rx_pause,
3536                                  mdev->profile.prof[priv->port].tx_ppp,
3537                                  mdev->profile.prof[priv->port].tx_pause);
3538
3539         err = register_netdev(dev);
3540         if (err) {
3541                 en_err(priv, "Netdev registration failed for port %d\n", port);
3542                 goto out;
3543         }
3544
3545         priv->registered = 1;
3546         devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3547                                   dev);
3548
3549         return 0;
3550
3551 err_free_tx:
3552         while (t--) {
3553                 kfree(priv->tx_ring[t]);
3554                 kfree(priv->tx_cq[t]);
3555         }
3556 out:
3557         mlx4_en_destroy_netdev(dev);
3558         return err;
3559 }
3560
3561 int mlx4_en_reset_config(struct net_device *dev,
3562                          struct hwtstamp_config ts_config,
3563                          netdev_features_t features)
3564 {
3565         struct mlx4_en_priv *priv = netdev_priv(dev);
3566         struct mlx4_en_dev *mdev = priv->mdev;
3567         struct mlx4_en_port_profile new_prof;
3568         struct mlx4_en_priv *tmp;
3569         int port_up = 0;
3570         int err = 0;
3571
3572         if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3573             priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3574             !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3575             !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3576                 return 0; /* Nothing to change */
3577
3578         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3579             (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3580             (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3581                 en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3582                 return -EINVAL;
3583         }
3584
3585         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3586         if (!tmp)
3587                 return -ENOMEM;
3588
3589         mutex_lock(&mdev->state_lock);
3590
3591         memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3592         memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3593
3594         err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3595         if (err)
3596                 goto out;
3597
3598         if (priv->port_up) {
3599                 port_up = 1;
3600                 mlx4_en_stop_port(dev, 1);
3601         }
3602
3603         en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
3604                 ts_config.rx_filter,
3605                 !!(features & NETIF_F_HW_VLAN_CTAG_RX));
3606
3607         mlx4_en_safe_replace_resources(priv, tmp);
3608
3609         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3610                 if (features & NETIF_F_HW_VLAN_CTAG_RX)
3611                         dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3612                 else
3613                         dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3614         } else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3615                 /* RX time-stamping is OFF, update the RX vlan offload
3616                  * to the latest wanted state
3617                  */
3618                 if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3619                         dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3620                 else
3621                         dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3622         }
3623
3624         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3625                 if (features & NETIF_F_RXFCS)
3626                         dev->features |= NETIF_F_RXFCS;
3627                 else
3628                         dev->features &= ~NETIF_F_RXFCS;
3629         }
3630
3631         /* RX vlan offload and RX time-stamping can't co-exist !
3632          * Regardless of the caller's choice,
3633          * Turn Off RX vlan offload in case of time-stamping is ON
3634          */
3635         if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3636                 if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3637                         en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3638                 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3639         }
3640
3641         if (port_up) {
3642                 err = mlx4_en_start_port(dev);
3643                 if (err)
3644                         en_err(priv, "Failed starting port\n");
3645         }
3646
3647 out:
3648         mutex_unlock(&mdev->state_lock);
3649         kfree(tmp);
3650         if (!err)
3651                 netdev_features_change(dev);
3652         return err;
3653 }