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)
  64                 return -EINVAL;
  65
  66         netdev_set_num_tc(dev, up);
  67
  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 static int __mlx4_en_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
  90                               struct tc_to_netdev *tc)
  91 {
  92         if (tc->type != TC_SETUP_MQPRIO)
  93                 return -EINVAL;
  94
  95         return mlx4_en_setup_tc(dev, tc->tc);
  96 }
  97
  98 #ifdef CONFIG_RFS_ACCEL
  99
 100 struct mlx4_en_filter {
 101         struct list_head next;
 102         struct work_struct work;
 103
 104         u8     ip_proto;
 105         __be32 src_ip;
 106         __be32 dst_ip;
 107         __be16 src_port;
 108         __be16 dst_port;
 109
 110         int rxq_index;
 111         struct mlx4_en_priv *priv;
 112         u32 flow_id;                    /* RFS infrastructure id */
 113         int id;                         /* mlx4_en driver id */
 114         u64 reg_id;                     /* Flow steering API id */
 115         u8 activated;                   /* Used to prevent expiry before filter
 116                                          * is attached
 117                                          */
 118         struct hlist_node filter_chain;
 119 };
 120
 121 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
 122
 123 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
 124 {
 125         switch (ip_proto) {
 126         case IPPROTO_UDP:
 127                 return MLX4_NET_TRANS_RULE_ID_UDP;
 128         case IPPROTO_TCP:
 129                 return MLX4_NET_TRANS_RULE_ID_TCP;
 130         default:
 131                 return MLX4_NET_TRANS_RULE_NUM;
 132         }
 133 };
 134
 135 /* Must not acquire state_lock, as its corresponding work_sync
 136  * is done under it.
 137  */
 138 static void mlx4_en_filter_work(struct work_struct *work)
 139 {
 140         struct mlx4_en_filter *filter = container_of(work,
 141                                                      struct mlx4_en_filter,
 142                                                      work);
 143         struct mlx4_en_priv *priv = filter->priv;
 144         struct mlx4_spec_list spec_tcp_udp = {
 145                 .id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
 146                 {
 147                         .tcp_udp = {
 148                                 .dst_port = filter->dst_port,
 149                                 .dst_port_msk = (__force __be16)-1,
 150                                 .src_port = filter->src_port,
 151                                 .src_port_msk = (__force __be16)-1,
 152                         },
 153                 },
 154         };
 155         struct mlx4_spec_list spec_ip = {
 156                 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
 157                 {
 158                         .ipv4 = {
 159                                 .dst_ip = filter->dst_ip,
 160                                 .dst_ip_msk = (__force __be32)-1,
 161                                 .src_ip = filter->src_ip,
 162                                 .src_ip_msk = (__force __be32)-1,
 163                         },
 164                 },
 165         };
 166         struct mlx4_spec_list spec_eth = {
 167                 .id = MLX4_NET_TRANS_RULE_ID_ETH,
 168         };
 169         struct mlx4_net_trans_rule rule = {
 170                 .list = LIST_HEAD_INIT(rule.list),
 171                 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
 172                 .exclusive = 1,
 173                 .allow_loopback = 1,
 174                 .promisc_mode = MLX4_FS_REGULAR,
 175                 .port = priv->port,
 176                 .priority = MLX4_DOMAIN_RFS,
 177         };
 178         int rc;
 179         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 180
 181         if (spec_tcp_udp.id >= MLX4_NET_TRANS_RULE_NUM) {
 182                 en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
 183                         filter->ip_proto);
 184                 goto ignore;
 185         }
 186         list_add_tail(&spec_eth.list, &rule.list);
 187         list_add_tail(&spec_ip.list, &rule.list);
 188         list_add_tail(&spec_tcp_udp.list, &rule.list);
 189
 190         rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
 191         memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
 192         memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 193
 194         filter->activated = 0;
 195
 196         if (filter->reg_id) {
 197                 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 198                 if (rc && rc != -ENOENT)
 199                         en_err(priv, "Error detaching flow. rc = %d\n", rc);
 200         }
 201
 202         rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
 203         if (rc)
 204                 en_err(priv, "Error attaching flow. err = %d\n", rc);
 205
 206 ignore:
 207         mlx4_en_filter_rfs_expire(priv);
 208
 209         filter->activated = 1;
 210 }
 211
 212 static inline struct hlist_head *
 213 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 214                    __be16 src_port, __be16 dst_port)
 215 {
 216         unsigned long l;
 217         int bucket_idx;
 218
 219         l = (__force unsigned long)src_port |
 220             ((__force unsigned long)dst_port << 2);
 221         l ^= (__force unsigned long)(src_ip ^ dst_ip);
 222
 223         bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
 224
 225         return &priv->filter_hash[bucket_idx];
 226 }
 227
 228 static struct mlx4_en_filter *
 229 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
 230                      __be32 dst_ip, u8 ip_proto, __be16 src_port,
 231                      __be16 dst_port, u32 flow_id)
 232 {
 233         struct mlx4_en_filter *filter = NULL;
 234
 235         filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
 236         if (!filter)
 237                 return NULL;
 238
 239         filter->priv = priv;
 240         filter->rxq_index = rxq_index;
 241         INIT_WORK(&filter->work, mlx4_en_filter_work);
 242
 243         filter->src_ip = src_ip;
 244         filter->dst_ip = dst_ip;
 245         filter->ip_proto = ip_proto;
 246         filter->src_port = src_port;
 247         filter->dst_port = dst_port;
 248
 249         filter->flow_id = flow_id;
 250
 251         filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
 252
 253         list_add_tail(&filter->next, &priv->filters);
 254         hlist_add_head(&filter->filter_chain,
 255                        filter_hash_bucket(priv, src_ip, dst_ip, src_port,
 256                                           dst_port));
 257
 258         return filter;
 259 }
 260
 261 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
 262 {
 263         struct mlx4_en_priv *priv = filter->priv;
 264         int rc;
 265
 266         list_del(&filter->next);
 267
 268         rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 269         if (rc && rc != -ENOENT)
 270                 en_err(priv, "Error detaching flow. rc = %d\n", rc);
 271
 272         kfree(filter);
 273 }
 274
 275 static inline struct mlx4_en_filter *
 276 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 277                     u8 ip_proto, __be16 src_port, __be16 dst_port)
 278 {
 279         struct mlx4_en_filter *filter;
 280         struct mlx4_en_filter *ret = NULL;
 281
 282         hlist_for_each_entry(filter,
 283                              filter_hash_bucket(priv, src_ip, dst_ip,
 284                                                 src_port, dst_port),
 285                              filter_chain) {
 286                 if (filter->src_ip == src_ip &&
 287                     filter->dst_ip == dst_ip &&
 288                     filter->ip_proto == ip_proto &&
 289                     filter->src_port == src_port &&
 290                     filter->dst_port == dst_port) {
 291                         ret = filter;
 292                         break;
 293                 }
 294         }
 295
 296         return ret;
 297 }
 298
 299 static int
 300 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
 301                    u16 rxq_index, u32 flow_id)
 302 {
 303         struct mlx4_en_priv *priv = netdev_priv(net_dev);
 304         struct mlx4_en_filter *filter;
 305         const struct iphdr *ip;
 306         const __be16 *ports;
 307         u8 ip_proto;
 308         __be32 src_ip;
 309         __be32 dst_ip;
 310         __be16 src_port;
 311         __be16 dst_port;
 312         int nhoff = skb_network_offset(skb);
 313         int ret = 0;
 314
 315         if (skb->protocol != htons(ETH_P_IP))
 316                 return -EPROTONOSUPPORT;
 317
 318         ip = (const struct iphdr *)(skb->data + nhoff);
 319         if (ip_is_fragment(ip))
 320                 return -EPROTONOSUPPORT;
 321
 322         if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
 323                 return -EPROTONOSUPPORT;
 324         ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
 325
 326         ip_proto = ip->protocol;
 327         src_ip = ip->saddr;
 328         dst_ip = ip->daddr;
 329         src_port = ports[0];
 330         dst_port = ports[1];
 331
 332         spin_lock_bh(&priv->filters_lock);
 333         filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
 334                                      src_port, dst_port);
 335         if (filter) {
 336                 if (filter->rxq_index == rxq_index)
 337                         goto out;
 338
 339                 filter->rxq_index = rxq_index;
 340         } else {
 341                 filter = mlx4_en_filter_alloc(priv, rxq_index,
 342                                               src_ip, dst_ip, ip_proto,
 343                                               src_port, dst_port, flow_id);
 344                 if (!filter) {
 345                         ret = -ENOMEM;
 346                         goto err;
 347                 }
 348         }
 349
 350         queue_work(priv->mdev->workqueue, &filter->work);
 351
 352 out:
 353         ret = filter->id;
 354 err:
 355         spin_unlock_bh(&priv->filters_lock);
 356
 357         return ret;
 358 }
 359
 360 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv)
 361 {
 362         struct mlx4_en_filter *filter, *tmp;
 363         LIST_HEAD(del_list);
 364
 365         spin_lock_bh(&priv->filters_lock);
 366         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 367                 list_move(&filter->next, &del_list);
 368                 hlist_del(&filter->filter_chain);
 369         }
 370         spin_unlock_bh(&priv->filters_lock);
 371
 372         list_for_each_entry_safe(filter, tmp, &del_list, next) {
 373                 cancel_work_sync(&filter->work);
 374                 mlx4_en_filter_free(filter);
 375         }
 376 }
 377
 378 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
 379 {
 380         struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
 381         LIST_HEAD(del_list);
 382         int i = 0;
 383
 384         spin_lock_bh(&priv->filters_lock);
 385         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 386                 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
 387                         break;
 388
 389                 if (filter->activated &&
 390                     !work_pending(&filter->work) &&
 391                     rps_may_expire_flow(priv->dev,
 392                                         filter->rxq_index, filter->flow_id,
 393                                         filter->id)) {
 394                         list_move(&filter->next, &del_list);
 395                         hlist_del(&filter->filter_chain);
 396                 } else
 397                         last_filter = filter;
 398
 399                 i++;
 400         }
 401
 402         if (last_filter && (&last_filter->next != priv->filters.next))
 403                 list_move(&priv->filters, &last_filter->next);
 404
 405         spin_unlock_bh(&priv->filters_lock);
 406
 407         list_for_each_entry_safe(filter, tmp, &del_list, next)
 408                 mlx4_en_filter_free(filter);
 409 }
 410 #endif
 411
 412 static int mlx4_en_vlan_rx_add_vid(struct net_device *dev,
 413                                    __be16 proto, u16 vid)
 414 {
 415         struct mlx4_en_priv *priv = netdev_priv(dev);
 416         struct mlx4_en_dev *mdev = priv->mdev;
 417         int err;
 418         int idx;
 419
 420         en_dbg(HW, priv, "adding VLAN:%d\n", vid);
 421
 422         set_bit(vid, priv->active_vlans);
 423
 424         /* Add VID to port VLAN filter */
 425         mutex_lock(&mdev->state_lock);
 426         if (mdev->device_up && priv->port_up) {
 427                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 428                 if (err) {
 429                         en_err(priv, "Failed configuring VLAN filter\n");
 430                         goto out;
 431                 }
 432         }
 433         err = mlx4_register_vlan(mdev->dev, priv->port, vid, &idx);
 434         if (err)
 435                 en_dbg(HW, priv, "Failed adding vlan %d\n", vid);
 436
 437 out:
 438         mutex_unlock(&mdev->state_lock);
 439         return err;
 440 }
 441
 442 static int mlx4_en_vlan_rx_kill_vid(struct net_device *dev,
 443                                     __be16 proto, u16 vid)
 444 {
 445         struct mlx4_en_priv *priv = netdev_priv(dev);
 446         struct mlx4_en_dev *mdev = priv->mdev;
 447         int err = 0;
 448
 449         en_dbg(HW, priv, "Killing VID:%d\n", vid);
 450
 451         clear_bit(vid, priv->active_vlans);
 452
 453         /* Remove VID from port VLAN filter */
 454         mutex_lock(&mdev->state_lock);
 455         mlx4_unregister_vlan(mdev->dev, priv->port, vid);
 456
 457         if (mdev->device_up && priv->port_up) {
 458                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 459                 if (err)
 460                         en_err(priv, "Failed configuring VLAN filter\n");
 461         }
 462         mutex_unlock(&mdev->state_lock);
 463
 464         return err;
 465 }
 466
 467 static void mlx4_en_u64_to_mac(unsigned char dst_mac[ETH_ALEN + 2], u64 src_mac)
 468 {
 469         int i;
 470         for (i = ETH_ALEN - 1; i >= 0; --i) {
 471                 dst_mac[i] = src_mac & 0xff;
 472                 src_mac >>= 8;
 473         }
 474         memset(&dst_mac[ETH_ALEN], 0, 2);
 475 }
 476
 477
 478 static int mlx4_en_tunnel_steer_add(struct mlx4_en_priv *priv, unsigned char *addr,
 479                                     int qpn, u64 *reg_id)
 480 {
 481         int err;
 482
 483         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
 484             priv->mdev->dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
 485                 return 0; /* do nothing */
 486
 487         err = mlx4_tunnel_steer_add(priv->mdev->dev, addr, priv->port, qpn,
 488                                     MLX4_DOMAIN_NIC, reg_id);
 489         if (err) {
 490                 en_err(priv, "failed to add vxlan steering rule, err %d\n", err);
 491                 return err;
 492         }
 493         en_dbg(DRV, priv, "added vxlan steering rule, mac %pM reg_id %llx\n", addr, *reg_id);
 494         return 0;
 495 }
 496
 497
 498 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
 499                                 unsigned char *mac, int *qpn, u64 *reg_id)
 500 {
 501         struct mlx4_en_dev *mdev = priv->mdev;
 502         struct mlx4_dev *dev = mdev->dev;
 503         int err;
 504
 505         switch (dev->caps.steering_mode) {
 506         case MLX4_STEERING_MODE_B0: {
 507                 struct mlx4_qp qp;
 508                 u8 gid[16] = {0};
 509
 510                 qp.qpn = *qpn;
 511                 memcpy(&gid[10], mac, ETH_ALEN);
 512                 gid[5] = priv->port;
 513
 514                 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
 515                 break;
 516         }
 517         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 518                 struct mlx4_spec_list spec_eth = { {NULL} };
 519                 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 520
 521                 struct mlx4_net_trans_rule rule = {
 522                         .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 523                         .exclusive = 0,
 524                         .allow_loopback = 1,
 525                         .promisc_mode = MLX4_FS_REGULAR,
 526                         .priority = MLX4_DOMAIN_NIC,
 527                 };
 528
 529                 rule.port = priv->port;
 530                 rule.qpn = *qpn;
 531                 INIT_LIST_HEAD(&rule.list);
 532
 533                 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
 534                 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
 535                 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 536                 list_add_tail(&spec_eth.list, &rule.list);
 537
 538                 err = mlx4_flow_attach(dev, &rule, reg_id);
 539                 break;
 540         }
 541         default:
 542                 return -EINVAL;
 543         }
 544         if (err)
 545                 en_warn(priv, "Failed Attaching Unicast\n");
 546
 547         return err;
 548 }
 549
 550 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
 551                                      unsigned char *mac, int qpn, u64 reg_id)
 552 {
 553         struct mlx4_en_dev *mdev = priv->mdev;
 554         struct mlx4_dev *dev = mdev->dev;
 555
 556         switch (dev->caps.steering_mode) {
 557         case MLX4_STEERING_MODE_B0: {
 558                 struct mlx4_qp qp;
 559                 u8 gid[16] = {0};
 560
 561                 qp.qpn = qpn;
 562                 memcpy(&gid[10], mac, ETH_ALEN);
 563                 gid[5] = priv->port;
 564
 565                 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
 566                 break;
 567         }
 568         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 569                 mlx4_flow_detach(dev, reg_id);
 570                 break;
 571         }
 572         default:
 573                 en_err(priv, "Invalid steering mode.\n");
 574         }
 575 }
 576
 577 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
 578 {
 579         struct mlx4_en_dev *mdev = priv->mdev;
 580         struct mlx4_dev *dev = mdev->dev;
 581         int index = 0;
 582         int err = 0;
 583         int *qpn = &priv->base_qpn;
 584         u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 585
 586         en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
 587                priv->dev->dev_addr);
 588         index = mlx4_register_mac(dev, priv->port, mac);
 589         if (index < 0) {
 590                 err = index;
 591                 en_err(priv, "Failed adding MAC: %pM\n",
 592                        priv->dev->dev_addr);
 593                 return err;
 594         }
 595
 596         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 597                 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
 598                 *qpn = base_qpn + index;
 599                 return 0;
 600         }
 601
 602         err = mlx4_qp_reserve_range(dev, 1, 1, qpn, MLX4_RESERVE_A0_QP);
 603         en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
 604         if (err) {
 605                 en_err(priv, "Failed to reserve qp for mac registration\n");
 606                 mlx4_unregister_mac(dev, priv->port, mac);
 607                 return err;
 608         }
 609
 610         return 0;
 611 }
 612
 613 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
 614 {
 615         struct mlx4_en_dev *mdev = priv->mdev;
 616         struct mlx4_dev *dev = mdev->dev;
 617         int qpn = priv->base_qpn;
 618
 619         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 620                 u64 mac = mlx4_mac_to_u64(priv->dev->dev_addr);
 621                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 622                        priv->dev->dev_addr);
 623                 mlx4_unregister_mac(dev, priv->port, mac);
 624         } else {
 625                 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
 626                        priv->port, qpn);
 627                 mlx4_qp_release_range(dev, qpn, 1);
 628                 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
 629         }
 630 }
 631
 632 static int mlx4_en_replace_mac(struct mlx4_en_priv *priv, int qpn,
 633                                unsigned char *new_mac, unsigned char *prev_mac)
 634 {
 635         struct mlx4_en_dev *mdev = priv->mdev;
 636         struct mlx4_dev *dev = mdev->dev;
 637         int err = 0;
 638         u64 new_mac_u64 = mlx4_mac_to_u64(new_mac);
 639
 640         if (dev->caps.steering_mode != MLX4_STEERING_MODE_A0) {
 641                 struct hlist_head *bucket;
 642                 unsigned int mac_hash;
 643                 struct mlx4_mac_entry *entry;
 644                 struct hlist_node *tmp;
 645                 u64 prev_mac_u64 = mlx4_mac_to_u64(prev_mac);
 646
 647                 bucket = &priv->mac_hash[prev_mac[MLX4_EN_MAC_HASH_IDX]];
 648                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
 649                         if (ether_addr_equal_64bits(entry->mac, prev_mac)) {
 650                                 mlx4_en_uc_steer_release(priv, entry->mac,
 651                                                          qpn, entry->reg_id);
 652                                 mlx4_unregister_mac(dev, priv->port,
 653                                                     prev_mac_u64);
 654                                 hlist_del_rcu(&entry->hlist);
 655                                 synchronize_rcu();
 656                                 memcpy(entry->mac, new_mac, ETH_ALEN);
 657                                 entry->reg_id = 0;
 658                                 mac_hash = new_mac[MLX4_EN_MAC_HASH_IDX];
 659                                 hlist_add_head_rcu(&entry->hlist,
 660                                                    &priv->mac_hash[mac_hash]);
 661                                 mlx4_register_mac(dev, priv->port, new_mac_u64);
 662                                 err = mlx4_en_uc_steer_add(priv, new_mac,
 663                                                            &qpn,
 664                                                            &entry->reg_id);
 665                                 if (err)
 666                                         return err;
 667                                 if (priv->tunnel_reg_id) {
 668                                         mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
 669                                         priv->tunnel_reg_id = 0;
 670                                 }
 671                                 err = mlx4_en_tunnel_steer_add(priv, new_mac, qpn,
 672                                                                &priv->tunnel_reg_id);
 673                                 return err;
 674                         }
 675                 }
 676                 return -EINVAL;
 677         }
 678
 679         return __mlx4_replace_mac(dev, priv->port, qpn, new_mac_u64);
 680 }
 681
 682 static int mlx4_en_do_set_mac(struct mlx4_en_priv *priv,
 683                               unsigned char new_mac[ETH_ALEN + 2])
 684 {
 685         int err = 0;
 686
 687         if (priv->port_up) {
 688                 /* Remove old MAC and insert the new one */
 689                 err = mlx4_en_replace_mac(priv, priv->base_qpn,
 690                                           new_mac, priv->current_mac);
 691                 if (err)
 692                         en_err(priv, "Failed changing HW MAC address\n");
 693         } else
 694                 en_dbg(HW, priv, "Port is down while registering mac, exiting...\n");
 695
 696         if (!err)
 697                 memcpy(priv->current_mac, new_mac, sizeof(priv->current_mac));
 698
 699         return err;
 700 }
 701
 702 static int mlx4_en_set_mac(struct net_device *dev, void *addr)
 703 {
 704         struct mlx4_en_priv *priv = netdev_priv(dev);
 705         struct mlx4_en_dev *mdev = priv->mdev;
 706         struct sockaddr *saddr = addr;
 707         unsigned char new_mac[ETH_ALEN + 2];
 708         int err;
 709
 710         if (!is_valid_ether_addr(saddr->sa_data))
 711                 return -EADDRNOTAVAIL;
 712
 713         mutex_lock(&mdev->state_lock);
 714         memcpy(new_mac, saddr->sa_data, ETH_ALEN);
 715         err = mlx4_en_do_set_mac(priv, new_mac);
 716         if (!err)
 717                 memcpy(dev->dev_addr, saddr->sa_data, ETH_ALEN);
 718         mutex_unlock(&mdev->state_lock);
 719
 720         return err;
 721 }
 722
 723 static void mlx4_en_clear_list(struct net_device *dev)
 724 {
 725         struct mlx4_en_priv *priv = netdev_priv(dev);
 726         struct mlx4_en_mc_list *tmp, *mc_to_del;
 727
 728         list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
 729                 list_del(&mc_to_del->list);
 730                 kfree(mc_to_del);
 731         }
 732 }
 733
 734 static void mlx4_en_cache_mclist(struct net_device *dev)
 735 {
 736         struct mlx4_en_priv *priv = netdev_priv(dev);
 737         struct netdev_hw_addr *ha;
 738         struct mlx4_en_mc_list *tmp;
 739
 740         mlx4_en_clear_list(dev);
 741         netdev_for_each_mc_addr(ha, dev) {
 742                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 743                 if (!tmp) {
 744                         mlx4_en_clear_list(dev);
 745                         return;
 746                 }
 747                 memcpy(tmp->addr, ha->addr, ETH_ALEN);
 748                 list_add_tail(&tmp->list, &priv->mc_list);
 749         }
 750 }
 751
 752 static void update_mclist_flags(struct mlx4_en_priv *priv,
 753                                 struct list_head *dst,
 754                                 struct list_head *src)
 755 {
 756         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 757         bool found;
 758
 759         /* Find all the entries that should be removed from dst,
 760          * These are the entries that are not found in src
 761          */
 762         list_for_each_entry(dst_tmp, dst, list) {
 763                 found = false;
 764                 list_for_each_entry(src_tmp, src, list) {
 765                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 766                                 found = true;
 767                                 break;
 768                         }
 769                 }
 770                 if (!found)
 771                         dst_tmp->action = MCLIST_REM;
 772         }
 773
 774         /* Add entries that exist in src but not in dst
 775          * mark them as need to add
 776          */
 777         list_for_each_entry(src_tmp, src, list) {
 778                 found = false;
 779                 list_for_each_entry(dst_tmp, dst, list) {
 780                         if (ether_addr_equal(dst_tmp->addr, src_tmp->addr)) {
 781                                 dst_tmp->action = MCLIST_NONE;
 782                                 found = true;
 783                                 break;
 784                         }
 785                 }
 786                 if (!found) {
 787                         new_mc = kmemdup(src_tmp,
 788                                          sizeof(struct mlx4_en_mc_list),
 789                                          GFP_KERNEL);
 790                         if (!new_mc)
 791                                 return;
 792
 793                         new_mc->action = MCLIST_ADD;
 794                         list_add_tail(&new_mc->list, dst);
 795                 }
 796         }
 797 }
 798
 799 static void mlx4_en_set_rx_mode(struct net_device *dev)
 800 {
 801         struct mlx4_en_priv *priv = netdev_priv(dev);
 802
 803         if (!priv->port_up)
 804                 return;
 805
 806         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 807 }
 808
 809 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 810                                      struct mlx4_en_dev *mdev)
 811 {
 812         int err = 0;
 813
 814         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 815                 if (netif_msg_rx_status(priv))
 816                         en_warn(priv, "Entering promiscuous mode\n");
 817                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 818
 819                 /* Enable promiscouos mode */
 820                 switch (mdev->dev->caps.steering_mode) {
 821                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 822                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 823                                                           priv->port,
 824                                                           priv->base_qpn,
 825                                                           MLX4_FS_ALL_DEFAULT);
 826                         if (err)
 827                                 en_err(priv, "Failed enabling promiscuous mode\n");
 828                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 829                         break;
 830
 831                 case MLX4_STEERING_MODE_B0:
 832                         err = mlx4_unicast_promisc_add(mdev->dev,
 833                                                        priv->base_qpn,
 834                                                        priv->port);
 835                         if (err)
 836                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 837
 838                         /* Add the default qp number as multicast
 839                          * promisc
 840                          */
 841                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 842                                 err = mlx4_multicast_promisc_add(mdev->dev,
 843                                                                  priv->base_qpn,
 844                                                                  priv->port);
 845                                 if (err)
 846                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 847                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 848                         }
 849                         break;
 850
 851                 case MLX4_STEERING_MODE_A0:
 852                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 853                                                      priv->port,
 854                                                      priv->base_qpn,
 855                                                      1);
 856                         if (err)
 857                                 en_err(priv, "Failed enabling promiscuous mode\n");
 858                         break;
 859                 }
 860
 861                 /* Disable port multicast filter (unconditionally) */
 862                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 863                                           0, MLX4_MCAST_DISABLE);
 864                 if (err)
 865                         en_err(priv, "Failed disabling multicast filter\n");
 866         }
 867 }
 868
 869 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 870                                        struct mlx4_en_dev *mdev)
 871 {
 872         int err = 0;
 873
 874         if (netif_msg_rx_status(priv))
 875                 en_warn(priv, "Leaving promiscuous mode\n");
 876         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 877
 878         /* Disable promiscouos mode */
 879         switch (mdev->dev->caps.steering_mode) {
 880         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 881                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 882                                                      priv->port,
 883                                                      MLX4_FS_ALL_DEFAULT);
 884                 if (err)
 885                         en_err(priv, "Failed disabling promiscuous mode\n");
 886                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 887                 break;
 888
 889         case MLX4_STEERING_MODE_B0:
 890                 err = mlx4_unicast_promisc_remove(mdev->dev,
 891                                                   priv->base_qpn,
 892                                                   priv->port);
 893                 if (err)
 894                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 895                 /* Disable Multicast promisc */
 896                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 897                         err = mlx4_multicast_promisc_remove(mdev->dev,
 898                                                             priv->base_qpn,
 899                                                             priv->port);
 900                         if (err)
 901                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 902                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 903                 }
 904                 break;
 905
 906         case MLX4_STEERING_MODE_A0:
 907                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 908                                              priv->port,
 909                                              priv->base_qpn, 0);
 910                 if (err)
 911                         en_err(priv, "Failed disabling promiscuous mode\n");
 912                 break;
 913         }
 914 }
 915
 916 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
 917                                  struct net_device *dev,
 918                                  struct mlx4_en_dev *mdev)
 919 {
 920         struct mlx4_en_mc_list *mclist, *tmp;
 921         u64 mcast_addr = 0;
 922         u8 mc_list[16] = {0};
 923         int err = 0;
 924
 925         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
 926         if (dev->flags & IFF_ALLMULTI) {
 927                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 928                                           0, MLX4_MCAST_DISABLE);
 929                 if (err)
 930                         en_err(priv, "Failed disabling multicast filter\n");
 931
 932                 /* Add the default qp number as multicast promisc */
 933                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 934                         switch (mdev->dev->caps.steering_mode) {
 935                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 936                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
 937                                                                   priv->port,
 938                                                                   priv->base_qpn,
 939                                                                   MLX4_FS_MC_DEFAULT);
 940                                 break;
 941
 942                         case MLX4_STEERING_MODE_B0:
 943                                 err = mlx4_multicast_promisc_add(mdev->dev,
 944                                                                  priv->base_qpn,
 945                                                                  priv->port);
 946                                 break;
 947
 948                         case MLX4_STEERING_MODE_A0:
 949                                 break;
 950                         }
 951                         if (err)
 952                                 en_err(priv, "Failed entering multicast promisc mode\n");
 953                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 954                 }
 955         } else {
 956                 /* Disable Multicast promisc */
 957                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 958                         switch (mdev->dev->caps.steering_mode) {
 959                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 960                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 961                                                                      priv->port,
 962                                                                      MLX4_FS_MC_DEFAULT);
 963                                 break;
 964
 965                         case MLX4_STEERING_MODE_B0:
 966                                 err = mlx4_multicast_promisc_remove(mdev->dev,
 967                                                                     priv->base_qpn,
 968                                                                     priv->port);
 969                                 break;
 970
 971                         case MLX4_STEERING_MODE_A0:
 972                                 break;
 973                         }
 974                         if (err)
 975                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 976                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 977                 }
 978
 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                 /* Flush mcast filter and init it with broadcast address */
 985                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
 986                                     1, MLX4_MCAST_CONFIG);
 987
 988                 /* Update multicast list - we cache all addresses so they won't
 989                  * change while HW is updated holding the command semaphor */
 990                 netif_addr_lock_bh(dev);
 991                 mlx4_en_cache_mclist(dev);
 992                 netif_addr_unlock_bh(dev);
 993                 list_for_each_entry(mclist, &priv->mc_list, list) {
 994                         mcast_addr = mlx4_mac_to_u64(mclist->addr);
 995                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
 996                                             mcast_addr, 0, MLX4_MCAST_CONFIG);
 997                 }
 998                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 999                                           0, MLX4_MCAST_ENABLE);
1000                 if (err)
1001                         en_err(priv, "Failed enabling multicast filter\n");
1002
1003                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
1004                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1005                         if (mclist->action == MCLIST_REM) {
1006                                 /* detach this address and delete from list */
1007                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1008                                 mc_list[5] = priv->port;
1009                                 err = mlx4_multicast_detach(mdev->dev,
1010                                                             &priv->rss_map.indir_qp,
1011                                                             mc_list,
1012                                                             MLX4_PROT_ETH,
1013                                                             mclist->reg_id);
1014                                 if (err)
1015                                         en_err(priv, "Fail to detach multicast address\n");
1016
1017                                 if (mclist->tunnel_reg_id) {
1018                                         err = mlx4_flow_detach(priv->mdev->dev, mclist->tunnel_reg_id);
1019                                         if (err)
1020                                                 en_err(priv, "Failed to detach multicast address\n");
1021                                 }
1022
1023                                 /* remove from list */
1024                                 list_del(&mclist->list);
1025                                 kfree(mclist);
1026                         } else if (mclist->action == MCLIST_ADD) {
1027                                 /* attach the address */
1028                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1029                                 /* needed for B0 steering support */
1030                                 mc_list[5] = priv->port;
1031                                 err = mlx4_multicast_attach(mdev->dev,
1032                                                             &priv->rss_map.indir_qp,
1033                                                             mc_list,
1034                                                             priv->port, 0,
1035                                                             MLX4_PROT_ETH,
1036                                                             &mclist->reg_id);
1037                                 if (err)
1038                                         en_err(priv, "Fail to attach multicast address\n");
1039
1040                                 err = mlx4_en_tunnel_steer_add(priv, &mc_list[10], priv->base_qpn,
1041                                                                &mclist->tunnel_reg_id);
1042                                 if (err)
1043                                         en_err(priv, "Failed to attach multicast address\n");
1044                         }
1045                 }
1046         }
1047 }
1048
1049 static void mlx4_en_do_uc_filter(struct mlx4_en_priv *priv,
1050                                  struct net_device *dev,
1051                                  struct mlx4_en_dev *mdev)
1052 {
1053         struct netdev_hw_addr *ha;
1054         struct mlx4_mac_entry *entry;
1055         struct hlist_node *tmp;
1056         bool found;
1057         u64 mac;
1058         int err = 0;
1059         struct hlist_head *bucket;
1060         unsigned int i;
1061         int removed = 0;
1062         u32 prev_flags;
1063
1064         /* Note that we do not need to protect our mac_hash traversal with rcu,
1065          * since all modification code is protected by mdev->state_lock
1066          */
1067
1068         /* find what to remove */
1069         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1070                 bucket = &priv->mac_hash[i];
1071                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1072                         found = false;
1073                         netdev_for_each_uc_addr(ha, dev) {
1074                                 if (ether_addr_equal_64bits(entry->mac,
1075                                                             ha->addr)) {
1076                                         found = true;
1077                                         break;
1078                                 }
1079                         }
1080
1081                         /* MAC address of the port is not in uc list */
1082                         if (ether_addr_equal_64bits(entry->mac,
1083                                                     priv->current_mac))
1084                                 found = true;
1085
1086                         if (!found) {
1087                                 mac = mlx4_mac_to_u64(entry->mac);
1088                                 mlx4_en_uc_steer_release(priv, entry->mac,
1089                                                          priv->base_qpn,
1090                                                          entry->reg_id);
1091                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1092
1093                                 hlist_del_rcu(&entry->hlist);
1094                                 kfree_rcu(entry, rcu);
1095                                 en_dbg(DRV, priv, "Removed MAC %pM on port:%d\n",
1096                                        entry->mac, priv->port);
1097                                 ++removed;
1098                         }
1099                 }
1100         }
1101
1102         /* if we didn't remove anything, there is no use in trying to add
1103          * again once we are in a forced promisc mode state
1104          */
1105         if ((priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) && 0 == removed)
1106                 return;
1107
1108         prev_flags = priv->flags;
1109         priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
1110
1111         /* find what to add */
1112         netdev_for_each_uc_addr(ha, dev) {
1113                 found = false;
1114                 bucket = &priv->mac_hash[ha->addr[MLX4_EN_MAC_HASH_IDX]];
1115                 hlist_for_each_entry(entry, bucket, hlist) {
1116                         if (ether_addr_equal_64bits(entry->mac, ha->addr)) {
1117                                 found = true;
1118                                 break;
1119                         }
1120                 }
1121
1122                 if (!found) {
1123                         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1124                         if (!entry) {
1125                                 en_err(priv, "Failed adding MAC %pM on port:%d (out of memory)\n",
1126                                        ha->addr, priv->port);
1127                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1128                                 break;
1129                         }
1130                         mac = mlx4_mac_to_u64(ha->addr);
1131                         memcpy(entry->mac, ha->addr, ETH_ALEN);
1132                         err = mlx4_register_mac(mdev->dev, priv->port, mac);
1133                         if (err < 0) {
1134                                 en_err(priv, "Failed registering MAC %pM on port %d: %d\n",
1135                                        ha->addr, priv->port, err);
1136                                 kfree(entry);
1137                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1138                                 break;
1139                         }
1140                         err = mlx4_en_uc_steer_add(priv, ha->addr,
1141                                                    &priv->base_qpn,
1142                                                    &entry->reg_id);
1143                         if (err) {
1144                                 en_err(priv, "Failed adding MAC %pM on port %d: %d\n",
1145                                        ha->addr, priv->port, err);
1146                                 mlx4_unregister_mac(mdev->dev, priv->port, mac);
1147                                 kfree(entry);
1148                                 priv->flags |= MLX4_EN_FLAG_FORCE_PROMISC;
1149                                 break;
1150                         } else {
1151                                 unsigned int mac_hash;
1152                                 en_dbg(DRV, priv, "Added MAC %pM on port:%d\n",
1153                                        ha->addr, priv->port);
1154                                 mac_hash = ha->addr[MLX4_EN_MAC_HASH_IDX];
1155                                 bucket = &priv->mac_hash[mac_hash];
1156                                 hlist_add_head_rcu(&entry->hlist, bucket);
1157                         }
1158                 }
1159         }
1160
1161         if (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1162                 en_warn(priv, "Forcing promiscuous mode on port:%d\n",
1163                         priv->port);
1164         } else if (prev_flags & MLX4_EN_FLAG_FORCE_PROMISC) {
1165                 en_warn(priv, "Stop forcing promiscuous mode on port:%d\n",
1166                         priv->port);
1167         }
1168 }
1169
1170 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
1171 {
1172         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1173                                                  rx_mode_task);
1174         struct mlx4_en_dev *mdev = priv->mdev;
1175         struct net_device *dev = priv->dev;
1176
1177         mutex_lock(&mdev->state_lock);
1178         if (!mdev->device_up) {
1179                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
1180                 goto out;
1181         }
1182         if (!priv->port_up) {
1183                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
1184                 goto out;
1185         }
1186
1187         if (!netif_carrier_ok(dev)) {
1188                 if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
1189                         if (priv->port_state.link_state) {
1190                                 priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
1191                                 netif_carrier_on(dev);
1192                                 en_dbg(LINK, priv, "Link Up\n");
1193                         }
1194                 }
1195         }
1196
1197         if (dev->priv_flags & IFF_UNICAST_FLT)
1198                 mlx4_en_do_uc_filter(priv, dev, mdev);
1199
1200         /* Promsicuous mode: disable all filters */
1201         if ((dev->flags & IFF_PROMISC) ||
1202             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
1203                 mlx4_en_set_promisc_mode(priv, mdev);
1204                 goto out;
1205         }
1206
1207         /* Not in promiscuous mode */
1208         if (priv->flags & MLX4_EN_FLAG_PROMISC)
1209                 mlx4_en_clear_promisc_mode(priv, mdev);
1210
1211         mlx4_en_do_multicast(priv, dev, mdev);
1212 out:
1213         mutex_unlock(&mdev->state_lock);
1214 }
1215
1216 #ifdef CONFIG_NET_POLL_CONTROLLER
1217 static void mlx4_en_netpoll(struct net_device *dev)
1218 {
1219         struct mlx4_en_priv *priv = netdev_priv(dev);
1220         struct mlx4_en_cq *cq;
1221         int i;
1222
1223         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1224                 cq = priv->tx_cq[TX][i];
1225                 napi_schedule(&cq->napi);
1226         }
1227 }
1228 #endif
1229
1230 static int mlx4_en_set_rss_steer_rules(struct mlx4_en_priv *priv)
1231 {
1232         u64 reg_id;
1233         int err = 0;
1234         int *qpn = &priv->base_qpn;
1235         struct mlx4_mac_entry *entry;
1236
1237         err = mlx4_en_uc_steer_add(priv, priv->dev->dev_addr, qpn, &reg_id);
1238         if (err)
1239                 return err;
1240
1241         err = mlx4_en_tunnel_steer_add(priv, priv->dev->dev_addr, *qpn,
1242                                        &priv->tunnel_reg_id);
1243         if (err)
1244                 goto tunnel_err;
1245
1246         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1247         if (!entry) {
1248                 err = -ENOMEM;
1249                 goto alloc_err;
1250         }
1251
1252         memcpy(entry->mac, priv->dev->dev_addr, sizeof(entry->mac));
1253         memcpy(priv->current_mac, entry->mac, sizeof(priv->current_mac));
1254         entry->reg_id = reg_id;
1255         hlist_add_head_rcu(&entry->hlist,
1256                            &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
1257
1258         return 0;
1259
1260 alloc_err:
1261         if (priv->tunnel_reg_id)
1262                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1263
1264 tunnel_err:
1265         mlx4_en_uc_steer_release(priv, priv->dev->dev_addr, *qpn, reg_id);
1266         return err;
1267 }
1268
1269 static void mlx4_en_delete_rss_steer_rules(struct mlx4_en_priv *priv)
1270 {
1271         u64 mac;
1272         unsigned int i;
1273         int qpn = priv->base_qpn;
1274         struct hlist_head *bucket;
1275         struct hlist_node *tmp;
1276         struct mlx4_mac_entry *entry;
1277
1278         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
1279                 bucket = &priv->mac_hash[i];
1280                 hlist_for_each_entry_safe(entry, tmp, bucket, hlist) {
1281                         mac = mlx4_mac_to_u64(entry->mac);
1282                         en_dbg(DRV, priv, "Registering MAC:%pM for deleting\n",
1283                                entry->mac);
1284                         mlx4_en_uc_steer_release(priv, entry->mac,
1285                                                  qpn, entry->reg_id);
1286
1287                         mlx4_unregister_mac(priv->mdev->dev, priv->port, mac);
1288                         hlist_del_rcu(&entry->hlist);
1289                         kfree_rcu(entry, rcu);
1290                 }
1291         }
1292
1293         if (priv->tunnel_reg_id) {
1294                 mlx4_flow_detach(priv->mdev->dev, priv->tunnel_reg_id);
1295                 priv->tunnel_reg_id = 0;
1296         }
1297 }
1298
1299 static void mlx4_en_tx_timeout(struct net_device *dev)
1300 {
1301         struct mlx4_en_priv *priv = netdev_priv(dev);
1302         struct mlx4_en_dev *mdev = priv->mdev;
1303         int i;
1304
1305         if (netif_msg_timer(priv))
1306                 en_warn(priv, "Tx timeout called on port:%d\n", priv->port);
1307
1308         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1309                 struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][i];
1310
1311                 if (!netif_tx_queue_stopped(netdev_get_tx_queue(dev, i)))
1312                         continue;
1313                 en_warn(priv, "TX timeout on queue: %d, QP: 0x%x, CQ: 0x%x, Cons: 0x%x, Prod: 0x%x\n",
1314                         i, tx_ring->qpn, tx_ring->sp_cqn,
1315                         tx_ring->cons, tx_ring->prod);
1316         }
1317
1318         priv->port_stats.tx_timeout++;
1319         en_dbg(DRV, priv, "Scheduling watchdog\n");
1320         queue_work(mdev->workqueue, &priv->watchdog_task);
1321 }
1322
1323
1324 static struct rtnl_link_stats64 *
1325 mlx4_en_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1326 {
1327         struct mlx4_en_priv *priv = netdev_priv(dev);
1328
1329         spin_lock_bh(&priv->stats_lock);
1330         mlx4_en_fold_software_stats(dev);
1331         netdev_stats_to_stats64(stats, &dev->stats);
1332         spin_unlock_bh(&priv->stats_lock);
1333
1334         return stats;
1335 }
1336
1337 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1338 {
1339         struct mlx4_en_cq *cq;
1340         int i, t;
1341
1342         /* If we haven't received a specific coalescing setting
1343          * (module param), we set the moderation parameters as follows:
1344          * - moder_cnt is set to the number of mtu sized packets to
1345          *   satisfy our coalescing target.
1346          * - moder_time is set to a fixed value.
1347          */
1348         priv->rx_frames = MLX4_EN_RX_COAL_TARGET;
1349         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1350         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1351         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1352         en_dbg(INTR, priv, "Default coalesing params for mtu:%d - rx_frames:%d rx_usecs:%d\n",
1353                priv->dev->mtu, priv->rx_frames, priv->rx_usecs);
1354
1355         /* Setup cq moderation params */
1356         for (i = 0; i < priv->rx_ring_num; i++) {
1357                 cq = priv->rx_cq[i];
1358                 cq->moder_cnt = priv->rx_frames;
1359                 cq->moder_time = priv->rx_usecs;
1360                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1361                 priv->last_moder_packets[i] = 0;
1362                 priv->last_moder_bytes[i] = 0;
1363         }
1364
1365         for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1366                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1367                         cq = priv->tx_cq[t][i];
1368                         cq->moder_cnt = priv->tx_frames;
1369                         cq->moder_time = priv->tx_usecs;
1370                 }
1371         }
1372
1373         /* Reset auto-moderation params */
1374         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1375         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1376         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1377         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1378         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1379         priv->adaptive_rx_coal = 1;
1380         priv->last_moder_jiffies = 0;
1381         priv->last_moder_tx_packets = 0;
1382 }
1383
1384 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1385 {
1386         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1387         struct mlx4_en_cq *cq;
1388         unsigned long packets;
1389         unsigned long rate;
1390         unsigned long avg_pkt_size;
1391         unsigned long rx_packets;
1392         unsigned long rx_bytes;
1393         unsigned long rx_pkt_diff;
1394         int moder_time;
1395         int ring, err;
1396
1397         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1398                 return;
1399
1400         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1401                 rx_packets = READ_ONCE(priv->rx_ring[ring]->packets);
1402                 rx_bytes = READ_ONCE(priv->rx_ring[ring]->bytes);
1403
1404                 rx_pkt_diff = ((unsigned long) (rx_packets -
1405                                 priv->last_moder_packets[ring]));
1406                 packets = rx_pkt_diff;
1407                 rate = packets * HZ / period;
1408                 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1409                                 priv->last_moder_bytes[ring])) / packets : 0;
1410
1411                 /* Apply auto-moderation only when packet rate
1412                  * exceeds a rate that it matters */
1413                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1414                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1415                         if (rate < priv->pkt_rate_low)
1416                                 moder_time = priv->rx_usecs_low;
1417                         else if (rate > priv->pkt_rate_high)
1418                                 moder_time = priv->rx_usecs_high;
1419                         else
1420                                 moder_time = (rate - priv->pkt_rate_low) *
1421                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1422                                         (priv->pkt_rate_high - priv->pkt_rate_low) +
1423                                         priv->rx_usecs_low;
1424                 } else {
1425                         moder_time = priv->rx_usecs_low;
1426                 }
1427
1428                 if (moder_time != priv->last_moder_time[ring]) {
1429                         priv->last_moder_time[ring] = moder_time;
1430                         cq = priv->rx_cq[ring];
1431                         cq->moder_time = moder_time;
1432                         cq->moder_cnt = priv->rx_frames;
1433                         err = mlx4_en_set_cq_moder(priv, cq);
1434                         if (err)
1435                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1436                                        ring);
1437                 }
1438                 priv->last_moder_packets[ring] = rx_packets;
1439                 priv->last_moder_bytes[ring] = rx_bytes;
1440         }
1441
1442         priv->last_moder_jiffies = jiffies;
1443 }
1444
1445 static void mlx4_en_do_get_stats(struct work_struct *work)
1446 {
1447         struct delayed_work *delay = to_delayed_work(work);
1448         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1449                                                  stats_task);
1450         struct mlx4_en_dev *mdev = priv->mdev;
1451         int err;
1452
1453         mutex_lock(&mdev->state_lock);
1454         if (mdev->device_up) {
1455                 if (priv->port_up) {
1456                         err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1457                         if (err)
1458                                 en_dbg(HW, priv, "Could not update stats\n");
1459
1460                         mlx4_en_auto_moderation(priv);
1461                 }
1462
1463                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1464         }
1465         if (mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port]) {
1466                 mlx4_en_do_set_mac(priv, priv->current_mac);
1467                 mdev->mac_removed[MLX4_MAX_PORTS + 1 - priv->port] = 0;
1468         }
1469         mutex_unlock(&mdev->state_lock);
1470 }
1471
1472 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1473  * periodically
1474  */
1475 static void mlx4_en_service_task(struct work_struct *work)
1476 {
1477         struct delayed_work *delay = to_delayed_work(work);
1478         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1479                                                  service_task);
1480         struct mlx4_en_dev *mdev = priv->mdev;
1481
1482         mutex_lock(&mdev->state_lock);
1483         if (mdev->device_up) {
1484                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
1485                         mlx4_en_ptp_overflow_check(mdev);
1486
1487                 mlx4_en_recover_from_oom(priv);
1488                 queue_delayed_work(mdev->workqueue, &priv->service_task,
1489                                    SERVICE_TASK_DELAY);
1490         }
1491         mutex_unlock(&mdev->state_lock);
1492 }
1493
1494 static void mlx4_en_linkstate(struct work_struct *work)
1495 {
1496         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1497                                                  linkstate_task);
1498         struct mlx4_en_dev *mdev = priv->mdev;
1499         int linkstate = priv->link_state;
1500
1501         mutex_lock(&mdev->state_lock);
1502         /* If observable port state changed set carrier state and
1503          * report to system log */
1504         if (priv->last_link_state != linkstate) {
1505                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1506                         en_info(priv, "Link Down\n");
1507                         netif_carrier_off(priv->dev);
1508                 } else {
1509                         en_info(priv, "Link Up\n");
1510                         netif_carrier_on(priv->dev);
1511                 }
1512         }
1513         priv->last_link_state = linkstate;
1514         mutex_unlock(&mdev->state_lock);
1515 }
1516
1517 static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1518 {
1519         struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
1520         int numa_node = priv->mdev->dev->numa_node;
1521
1522         if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
1523                 return -ENOMEM;
1524
1525         cpumask_set_cpu(cpumask_local_spread(ring_idx, numa_node),
1526                         ring->affinity_mask);
1527         return 0;
1528 }
1529
1530 static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
1531 {
1532         free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
1533 }
1534
1535 static void mlx4_en_init_recycle_ring(struct mlx4_en_priv *priv,
1536                                       int tx_ring_idx)
1537 {
1538         struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX_XDP][tx_ring_idx];
1539         int rr_index = tx_ring_idx;
1540
1541         tx_ring->free_tx_desc = mlx4_en_recycle_tx_desc;
1542         tx_ring->recycle_ring = priv->rx_ring[rr_index];
1543         en_dbg(DRV, priv, "Set tx_ring[%d][%d]->recycle_ring = rx_ring[%d]\n",
1544                TX_XDP, tx_ring_idx, rr_index);
1545 }
1546
1547 int mlx4_en_start_port(struct net_device *dev)
1548 {
1549         struct mlx4_en_priv *priv = netdev_priv(dev);
1550         struct mlx4_en_dev *mdev = priv->mdev;
1551         struct mlx4_en_cq *cq;
1552         struct mlx4_en_tx_ring *tx_ring;
1553         int rx_index = 0;
1554         int err = 0;
1555         int i, t;
1556         int j;
1557         u8 mc_list[16] = {0};
1558
1559         if (priv->port_up) {
1560                 en_dbg(DRV, priv, "start port called while port already up\n");
1561                 return 0;
1562         }
1563
1564         INIT_LIST_HEAD(&priv->mc_list);
1565         INIT_LIST_HEAD(&priv->curr_list);
1566         INIT_LIST_HEAD(&priv->ethtool_list);
1567         memset(&priv->ethtool_rules[0], 0,
1568                sizeof(struct ethtool_flow_id) * MAX_NUM_OF_FS_RULES);
1569
1570         /* Calculate Rx buf size */
1571         dev->mtu = min(dev->mtu, priv->max_mtu);
1572         mlx4_en_calc_rx_buf(dev);
1573         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_skb_size);
1574
1575         /* Configure rx cq's and rings */
1576         err = mlx4_en_activate_rx_rings(priv);
1577         if (err) {
1578                 en_err(priv, "Failed to activate RX rings\n");
1579                 return err;
1580         }
1581         for (i = 0; i < priv->rx_ring_num; i++) {
1582                 cq = priv->rx_cq[i];
1583
1584                 err = mlx4_en_init_affinity_hint(priv, i);
1585                 if (err) {
1586                         en_err(priv, "Failed preparing IRQ affinity hint\n");
1587                         goto cq_err;
1588                 }
1589
1590                 err = mlx4_en_activate_cq(priv, cq, i);
1591                 if (err) {
1592                         en_err(priv, "Failed activating Rx CQ\n");
1593                         mlx4_en_free_affinity_hint(priv, i);
1594                         goto cq_err;
1595                 }
1596
1597                 for (j = 0; j < cq->size; j++) {
1598                         struct mlx4_cqe *cqe = NULL;
1599
1600                         cqe = mlx4_en_get_cqe(cq->buf, j, priv->cqe_size) +
1601                               priv->cqe_factor;
1602                         cqe->owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1603                 }
1604
1605                 err = mlx4_en_set_cq_moder(priv, cq);
1606                 if (err) {
1607                         en_err(priv, "Failed setting cq moderation parameters\n");
1608                         mlx4_en_deactivate_cq(priv, cq);
1609                         mlx4_en_free_affinity_hint(priv, i);
1610                         goto cq_err;
1611                 }
1612                 mlx4_en_arm_cq(priv, cq);
1613                 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1614                 ++rx_index;
1615         }
1616
1617         /* Set qp number */
1618         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1619         err = mlx4_en_get_qp(priv);
1620         if (err) {
1621                 en_err(priv, "Failed getting eth qp\n");
1622                 goto cq_err;
1623         }
1624         mdev->mac_removed[priv->port] = 0;
1625
1626         priv->counter_index =
1627                         mlx4_get_default_counter_index(mdev->dev, priv->port);
1628
1629         err = mlx4_en_config_rss_steer(priv);
1630         if (err) {
1631                 en_err(priv, "Failed configuring rss steering\n");
1632                 goto mac_err;
1633         }
1634
1635         err = mlx4_en_create_drop_qp(priv);
1636         if (err)
1637                 goto rss_err;
1638
1639         /* Configure tx cq's and rings */
1640         for (t = 0 ; t < MLX4_EN_NUM_TX_TYPES; t++) {
1641                 u8 num_tx_rings_p_up = t == TX ?
1642                         priv->num_tx_rings_p_up : priv->tx_ring_num[t];
1643
1644                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1645                         /* Configure cq */
1646                         cq = priv->tx_cq[t][i];
1647                         err = mlx4_en_activate_cq(priv, cq, i);
1648                         if (err) {
1649                                 en_err(priv, "Failed allocating Tx CQ\n");
1650                                 goto tx_err;
1651                         }
1652                         err = mlx4_en_set_cq_moder(priv, cq);
1653                         if (err) {
1654                                 en_err(priv, "Failed setting cq moderation parameters\n");
1655                                 mlx4_en_deactivate_cq(priv, cq);
1656                                 goto tx_err;
1657                         }
1658                         en_dbg(DRV, priv,
1659                                "Resetting index of collapsed CQ:%d to -1\n", i);
1660                         cq->buf->wqe_index = cpu_to_be16(0xffff);
1661
1662                         /* Configure ring */
1663                         tx_ring = priv->tx_ring[t][i];
1664                         err = mlx4_en_activate_tx_ring(priv, tx_ring,
1665                                                        cq->mcq.cqn,
1666                                                        i / num_tx_rings_p_up);
1667                         if (err) {
1668                                 en_err(priv, "Failed allocating Tx ring\n");
1669                                 mlx4_en_deactivate_cq(priv, cq);
1670                                 goto tx_err;
1671                         }
1672                         if (t != TX_XDP) {
1673                                 tx_ring->tx_queue = netdev_get_tx_queue(dev, i);
1674                                 tx_ring->recycle_ring = NULL;
1675                         } else {
1676                                 mlx4_en_init_recycle_ring(priv, i);
1677                         }
1678
1679                         /* Arm CQ for TX completions */
1680                         mlx4_en_arm_cq(priv, cq);
1681
1682                         /* Set initial ownership of all Tx TXBBs to SW (1) */
1683                         for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1684                                 *((u32 *)(tx_ring->buf + j)) = 0xffffffff;
1685                 }
1686         }
1687
1688         /* Configure port */
1689         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1690                                     priv->rx_skb_size + ETH_FCS_LEN,
1691                                     priv->prof->tx_pause,
1692                                     priv->prof->tx_ppp,
1693                                     priv->prof->rx_pause,
1694                                     priv->prof->rx_ppp);
1695         if (err) {
1696                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1697                        priv->port, err);
1698                 goto tx_err;
1699         }
1700         /* Set default qp number */
1701         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1702         if (err) {
1703                 en_err(priv, "Failed setting default qp numbers\n");
1704                 goto tx_err;
1705         }
1706
1707         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
1708                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
1709                 if (err) {
1710                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
1711                                err);
1712                         goto tx_err;
1713                 }
1714         }
1715
1716         /* Init port */
1717         en_dbg(HW, priv, "Initializing port\n");
1718         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1719         if (err) {
1720                 en_err(priv, "Failed Initializing port\n");
1721                 goto tx_err;
1722         }
1723
1724         /* Set Unicast and VXLAN steering rules */
1725         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0 &&
1726             mlx4_en_set_rss_steer_rules(priv))
1727                 mlx4_warn(mdev, "Failed setting steering rules\n");
1728
1729         /* Attach rx QP to bradcast address */
1730         eth_broadcast_addr(&mc_list[10]);
1731         mc_list[5] = priv->port; /* needed for B0 steering support */
1732         if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1733                                   priv->port, 0, MLX4_PROT_ETH,
1734                                   &priv->broadcast_id))
1735                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1736
1737         /* Must redo promiscuous mode setup. */
1738         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1739
1740         /* Schedule multicast task to populate multicast list */
1741         queue_work(mdev->workqueue, &priv->rx_mode_task);
1742
1743         if (priv->mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
1744                 udp_tunnel_get_rx_info(dev);
1745
1746         priv->port_up = true;
1747
1748         /* Process all completions if exist to prevent
1749          * the queues freezing if they are full
1750          */
1751         for (i = 0; i < priv->rx_ring_num; i++) {
1752                 local_bh_disable();
1753                 napi_schedule(&priv->rx_cq[i]->napi);
1754                 local_bh_enable();
1755         }
1756
1757         netif_tx_start_all_queues(dev);
1758         netif_device_attach(dev);
1759
1760         return 0;
1761
1762 tx_err:
1763         if (t == MLX4_EN_NUM_TX_TYPES) {
1764                 t--;
1765                 i = priv->tx_ring_num[t];
1766         }
1767         while (t >= 0) {
1768                 while (i--) {
1769                         mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1770                         mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1771                 }
1772                 if (!t--)
1773                         break;
1774                 i = priv->tx_ring_num[t];
1775         }
1776         mlx4_en_destroy_drop_qp(priv);
1777 rss_err:
1778         mlx4_en_release_rss_steer(priv);
1779 mac_err:
1780         mlx4_en_put_qp(priv);
1781 cq_err:
1782         while (rx_index--) {
1783                 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1784                 mlx4_en_free_affinity_hint(priv, rx_index);
1785         }
1786         for (i = 0; i < priv->rx_ring_num; i++)
1787                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1788
1789         return err; /* need to close devices */
1790 }
1791
1792
1793 void mlx4_en_stop_port(struct net_device *dev, int detach)
1794 {
1795         struct mlx4_en_priv *priv = netdev_priv(dev);
1796         struct mlx4_en_dev *mdev = priv->mdev;
1797         struct mlx4_en_mc_list *mclist, *tmp;
1798         struct ethtool_flow_id *flow, *tmp_flow;
1799         int i, t;
1800         u8 mc_list[16] = {0};
1801
1802         if (!priv->port_up) {
1803                 en_dbg(DRV, priv, "stop port called while port already down\n");
1804                 return;
1805         }
1806
1807         /* close port*/
1808         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1809
1810         /* Synchronize with tx routine */
1811         netif_tx_lock_bh(dev);
1812         if (detach)
1813                 netif_device_detach(dev);
1814         netif_tx_stop_all_queues(dev);
1815         netif_tx_unlock_bh(dev);
1816
1817         netif_tx_disable(dev);
1818
1819         spin_lock_bh(&priv->stats_lock);
1820         mlx4_en_fold_software_stats(dev);
1821         /* Set port as not active */
1822         priv->port_up = false;
1823         spin_unlock_bh(&priv->stats_lock);
1824
1825         priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
1826
1827         /* Promsicuous mode */
1828         if (mdev->dev->caps.steering_mode ==
1829             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1830                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1831                                  MLX4_EN_FLAG_MC_PROMISC);
1832                 mlx4_flow_steer_promisc_remove(mdev->dev,
1833                                                priv->port,
1834                                                MLX4_FS_ALL_DEFAULT);
1835                 mlx4_flow_steer_promisc_remove(mdev->dev,
1836                                                priv->port,
1837                                                MLX4_FS_MC_DEFAULT);
1838         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1839                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1840
1841                 /* Disable promiscouos mode */
1842                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1843                                             priv->port);
1844
1845                 /* Disable Multicast promisc */
1846                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1847                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1848                                                       priv->port);
1849                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1850                 }
1851         }
1852
1853         /* Detach All multicasts */
1854         eth_broadcast_addr(&mc_list[10]);
1855         mc_list[5] = priv->port; /* needed for B0 steering support */
1856         mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1857                               MLX4_PROT_ETH, priv->broadcast_id);
1858         list_for_each_entry(mclist, &priv->curr_list, list) {
1859                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1860                 mc_list[5] = priv->port;
1861                 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1862                                       mc_list, MLX4_PROT_ETH, mclist->reg_id);
1863                 if (mclist->tunnel_reg_id)
1864                         mlx4_flow_detach(mdev->dev, mclist->tunnel_reg_id);
1865         }
1866         mlx4_en_clear_list(dev);
1867         list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1868                 list_del(&mclist->list);
1869                 kfree(mclist);
1870         }
1871
1872         /* Flush multicast filter */
1873         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1874
1875         /* Remove flow steering rules for the port*/
1876         if (mdev->dev->caps.steering_mode ==
1877             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1878                 ASSERT_RTNL();
1879                 list_for_each_entry_safe(flow, tmp_flow,
1880                                          &priv->ethtool_list, list) {
1881                         mlx4_flow_detach(mdev->dev, flow->id);
1882                         list_del(&flow->list);
1883                 }
1884         }
1885
1886         mlx4_en_destroy_drop_qp(priv);
1887
1888         /* Free TX Rings */
1889         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
1890                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
1891                         mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[t][i]);
1892                         mlx4_en_deactivate_cq(priv, priv->tx_cq[t][i]);
1893                 }
1894         }
1895         msleep(10);
1896
1897         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++)
1898                 for (i = 0; i < priv->tx_ring_num[t]; i++)
1899                         mlx4_en_free_tx_buf(dev, priv->tx_ring[t][i]);
1900
1901         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
1902                 mlx4_en_delete_rss_steer_rules(priv);
1903
1904         /* Free RSS qps */
1905         mlx4_en_release_rss_steer(priv);
1906
1907         /* Unregister Mac address for the port */
1908         mlx4_en_put_qp(priv);
1909         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_REASSIGN_MAC_EN))
1910                 mdev->mac_removed[priv->port] = 1;
1911
1912         /* Free RX Rings */
1913         for (i = 0; i < priv->rx_ring_num; i++) {
1914                 struct mlx4_en_cq *cq = priv->rx_cq[i];
1915
1916                 napi_synchronize(&cq->napi);
1917                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1918                 mlx4_en_deactivate_cq(priv, cq);
1919
1920                 mlx4_en_free_affinity_hint(priv, i);
1921         }
1922 }
1923
1924 static void mlx4_en_restart(struct work_struct *work)
1925 {
1926         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1927                                                  watchdog_task);
1928         struct mlx4_en_dev *mdev = priv->mdev;
1929         struct net_device *dev = priv->dev;
1930
1931         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1932
1933         rtnl_lock();
1934         mutex_lock(&mdev->state_lock);
1935         if (priv->port_up) {
1936                 mlx4_en_stop_port(dev, 1);
1937                 if (mlx4_en_start_port(dev))
1938                         en_err(priv, "Failed restarting port %d\n", priv->port);
1939         }
1940         mutex_unlock(&mdev->state_lock);
1941         rtnl_unlock();
1942 }
1943
1944 static void mlx4_en_clear_stats(struct net_device *dev)
1945 {
1946         struct mlx4_en_priv *priv = netdev_priv(dev);
1947         struct mlx4_en_dev *mdev = priv->mdev;
1948         struct mlx4_en_tx_ring **tx_ring;
1949         int i;
1950
1951         if (!mlx4_is_slave(mdev->dev))
1952                 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1953                         en_dbg(HW, priv, "Failed dumping statistics\n");
1954
1955         memset(&priv->pstats, 0, sizeof(priv->pstats));
1956         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1957         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1958         memset(&priv->rx_flowstats, 0, sizeof(priv->rx_flowstats));
1959         memset(&priv->tx_flowstats, 0, sizeof(priv->tx_flowstats));
1960         memset(&priv->rx_priority_flowstats, 0,
1961                sizeof(priv->rx_priority_flowstats));
1962         memset(&priv->tx_priority_flowstats, 0,
1963                sizeof(priv->tx_priority_flowstats));
1964         memset(&priv->pf_stats, 0, sizeof(priv->pf_stats));
1965
1966         tx_ring = priv->tx_ring[TX];
1967         for (i = 0; i < priv->tx_ring_num[TX]; i++) {
1968                 tx_ring[i]->bytes = 0;
1969                 tx_ring[i]->packets = 0;
1970                 tx_ring[i]->tx_csum = 0;
1971                 tx_ring[i]->tx_dropped = 0;
1972                 tx_ring[i]->queue_stopped = 0;
1973                 tx_ring[i]->wake_queue = 0;
1974                 tx_ring[i]->tso_packets = 0;
1975                 tx_ring[i]->xmit_more = 0;
1976         }
1977         for (i = 0; i < priv->rx_ring_num; i++) {
1978                 priv->rx_ring[i]->bytes = 0;
1979                 priv->rx_ring[i]->packets = 0;
1980                 priv->rx_ring[i]->csum_ok = 0;
1981                 priv->rx_ring[i]->csum_none = 0;
1982                 priv->rx_ring[i]->csum_complete = 0;
1983         }
1984 }
1985
1986 static int mlx4_en_open(struct net_device *dev)
1987 {
1988         struct mlx4_en_priv *priv = netdev_priv(dev);
1989         struct mlx4_en_dev *mdev = priv->mdev;
1990         int err = 0;
1991
1992         mutex_lock(&mdev->state_lock);
1993
1994         if (!mdev->device_up) {
1995                 en_err(priv, "Cannot open - device down/disabled\n");
1996                 err = -EBUSY;
1997                 goto out;
1998         }
1999
2000         /* Reset HW statistics and SW counters */
2001         mlx4_en_clear_stats(dev);
2002
2003         err = mlx4_en_start_port(dev);
2004         if (err)
2005                 en_err(priv, "Failed starting port:%d\n", priv->port);
2006
2007 out:
2008         mutex_unlock(&mdev->state_lock);
2009         return err;
2010 }
2011
2012
2013 static int mlx4_en_close(struct net_device *dev)
2014 {
2015         struct mlx4_en_priv *priv = netdev_priv(dev);
2016         struct mlx4_en_dev *mdev = priv->mdev;
2017
2018         en_dbg(IFDOWN, priv, "Close port called\n");
2019
2020         mutex_lock(&mdev->state_lock);
2021
2022         mlx4_en_stop_port(dev, 0);
2023         netif_carrier_off(dev);
2024
2025         mutex_unlock(&mdev->state_lock);
2026         return 0;
2027 }
2028
2029 static void mlx4_en_free_resources(struct mlx4_en_priv *priv)
2030 {
2031         int i, t;
2032
2033 #ifdef CONFIG_RFS_ACCEL
2034         priv->dev->rx_cpu_rmap = NULL;
2035 #endif
2036
2037         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2038                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2039                         if (priv->tx_ring[t] && priv->tx_ring[t][i])
2040                                 mlx4_en_destroy_tx_ring(priv,
2041                                                         &priv->tx_ring[t][i]);
2042                         if (priv->tx_cq[t] && priv->tx_cq[t][i])
2043                                 mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2044                 }
2045                 kfree(priv->tx_ring[t]);
2046                 kfree(priv->tx_cq[t]);
2047         }
2048
2049         for (i = 0; i < priv->rx_ring_num; i++) {
2050                 if (priv->rx_ring[i])
2051                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2052                                 priv->prof->rx_ring_size, priv->stride);
2053                 if (priv->rx_cq[i])
2054                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2055         }
2056
2057 }
2058
2059 static int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
2060 {
2061         struct mlx4_en_port_profile *prof = priv->prof;
2062         int i, t;
2063         int node;
2064
2065         /* Create tx Rings */
2066         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2067                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2068                         node = cpu_to_node(i % num_online_cpus());
2069                         if (mlx4_en_create_cq(priv, &priv->tx_cq[t][i],
2070                                               prof->tx_ring_size, i, t, node))
2071                                 goto err;
2072
2073                         if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[t][i],
2074                                                    prof->tx_ring_size,
2075                                                    TXBB_SIZE, node, i))
2076                                 goto err;
2077                 }
2078         }
2079
2080         /* Create rx Rings */
2081         for (i = 0; i < priv->rx_ring_num; i++) {
2082                 node = cpu_to_node(i % num_online_cpus());
2083                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
2084                                       prof->rx_ring_size, i, RX, node))
2085                         goto err;
2086
2087                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
2088                                            prof->rx_ring_size, priv->stride,
2089                                            node))
2090                         goto err;
2091         }
2092
2093 #ifdef CONFIG_RFS_ACCEL
2094         priv->dev->rx_cpu_rmap = mlx4_get_cpu_rmap(priv->mdev->dev, priv->port);
2095 #endif
2096
2097         return 0;
2098
2099 err:
2100         en_err(priv, "Failed to allocate NIC resources\n");
2101         for (i = 0; i < priv->rx_ring_num; i++) {
2102                 if (priv->rx_ring[i])
2103                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
2104                                                 prof->rx_ring_size,
2105                                                 priv->stride);
2106                 if (priv->rx_cq[i])
2107                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
2108         }
2109         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2110                 for (i = 0; i < priv->tx_ring_num[t]; i++) {
2111                         if (priv->tx_ring[t][i])
2112                                 mlx4_en_destroy_tx_ring(priv,
2113                                                         &priv->tx_ring[t][i]);
2114                         if (priv->tx_cq[t][i])
2115                                 mlx4_en_destroy_cq(priv, &priv->tx_cq[t][i]);
2116                 }
2117         }
2118         return -ENOMEM;
2119 }
2120
2121
2122 static int mlx4_en_copy_priv(struct mlx4_en_priv *dst,
2123                              struct mlx4_en_priv *src,
2124                              struct mlx4_en_port_profile *prof)
2125 {
2126         int t;
2127
2128         memcpy(&dst->hwtstamp_config, &prof->hwtstamp_config,
2129                sizeof(dst->hwtstamp_config));
2130         dst->num_tx_rings_p_up = src->mdev->profile.num_tx_rings_p_up;
2131         dst->rx_ring_num = prof->rx_ring_num;
2132         dst->flags = prof->flags;
2133         dst->mdev = src->mdev;
2134         dst->port = src->port;
2135         dst->dev = src->dev;
2136         dst->prof = prof;
2137         dst->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2138                                          DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
2139
2140         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2141                 dst->tx_ring_num[t] = prof->tx_ring_num[t];
2142                 if (!dst->tx_ring_num[t])
2143                         continue;
2144
2145                 dst->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
2146                                           MAX_TX_RINGS, GFP_KERNEL);
2147                 if (!dst->tx_ring[t])
2148                         goto err_free_tx;
2149
2150                 dst->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
2151                                         MAX_TX_RINGS, GFP_KERNEL);
2152                 if (!dst->tx_cq[t]) {
2153                         kfree(dst->tx_ring[t]);
2154                         goto err_free_tx;
2155                 }
2156         }
2157
2158         return 0;
2159
2160 err_free_tx:
2161         while (t--) {
2162                 kfree(dst->tx_ring[t]);
2163                 kfree(dst->tx_cq[t]);
2164         }
2165         return -ENOMEM;
2166 }
2167
2168 static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
2169                                 struct mlx4_en_priv *src)
2170 {
2171         int t;
2172         memcpy(dst->rx_ring, src->rx_ring,
2173                sizeof(struct mlx4_en_rx_ring *) * src->rx_ring_num);
2174         memcpy(dst->rx_cq, src->rx_cq,
2175                sizeof(struct mlx4_en_cq *) * src->rx_ring_num);
2176         memcpy(&dst->hwtstamp_config, &src->hwtstamp_config,
2177                sizeof(dst->hwtstamp_config));
2178         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2179                 dst->tx_ring_num[t] = src->tx_ring_num[t];
2180                 dst->tx_ring[t] = src->tx_ring[t];
2181                 dst->tx_cq[t] = src->tx_cq[t];
2182         }
2183         dst->rx_ring_num = src->rx_ring_num;
2184         memcpy(dst->prof, src->prof, sizeof(struct mlx4_en_port_profile));
2185 }
2186
2187 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
2188                                 struct mlx4_en_priv *tmp,
2189                                 struct mlx4_en_port_profile *prof,
2190                                 bool carry_xdp_prog)
2191 {
2192         struct bpf_prog *xdp_prog;
2193         int i, t;
2194
2195         mlx4_en_copy_priv(tmp, priv, prof);
2196
2197         if (mlx4_en_alloc_resources(tmp)) {
2198                 en_warn(priv,
2199                         "%s: Resource allocation failed, using previous configuration\n",
2200                         __func__);
2201                 for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
2202                         kfree(tmp->tx_ring[t]);
2203                         kfree(tmp->tx_cq[t]);
2204                 }
2205                 return -ENOMEM;
2206         }
2207
2208         /* All rx_rings has the same xdp_prog.  Pick the first one. */
2209         xdp_prog = rcu_dereference_protected(
2210                 priv->rx_ring[0]->xdp_prog,
2211                 lockdep_is_held(&priv->mdev->state_lock));
2212
2213         if (xdp_prog && carry_xdp_prog) {
2214                 xdp_prog = bpf_prog_add(xdp_prog, tmp->rx_ring_num);
2215                 if (IS_ERR(xdp_prog)) {
2216                         mlx4_en_free_resources(tmp);
2217                         return PTR_ERR(xdp_prog);
2218                 }
2219                 for (i = 0; i < tmp->rx_ring_num; i++)
2220                         rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
2221                                            xdp_prog);
2222         }
2223
2224         return 0;
2225 }
2226
2227 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
2228                                     struct mlx4_en_priv *tmp)
2229 {
2230         mlx4_en_free_resources(priv);
2231         mlx4_en_update_priv(priv, tmp);
2232 }
2233
2234 void mlx4_en_destroy_netdev(struct net_device *dev)
2235 {
2236         struct mlx4_en_priv *priv = netdev_priv(dev);
2237         struct mlx4_en_dev *mdev = priv->mdev;
2238
2239         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
2240
2241         /* Unregister device - this will close the port if it was up */
2242         if (priv->registered) {
2243                 devlink_port_type_clear(mlx4_get_devlink_port(mdev->dev,
2244                                                               priv->port));
2245                 unregister_netdev(dev);
2246         }
2247
2248         if (priv->allocated)
2249                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
2250
2251         cancel_delayed_work(&priv->stats_task);
2252         cancel_delayed_work(&priv->service_task);
2253         /* flush any pending task for this netdev */
2254         flush_workqueue(mdev->workqueue);
2255
2256         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2257                 mlx4_en_remove_timestamp(mdev);
2258
2259         /* Detach the netdev so tasks would not attempt to access it */
2260         mutex_lock(&mdev->state_lock);
2261         mdev->pndev[priv->port] = NULL;
2262         mdev->upper[priv->port] = NULL;
2263
2264 #ifdef CONFIG_RFS_ACCEL
2265         mlx4_en_cleanup_filters(priv);
2266 #endif
2267
2268         mlx4_en_free_resources(priv);
2269         mutex_unlock(&mdev->state_lock);
2270
2271         free_netdev(dev);
2272 }
2273
2274 static bool mlx4_en_check_xdp_mtu(struct net_device *dev, int mtu)
2275 {
2276         struct mlx4_en_priv *priv = netdev_priv(dev);
2277
2278         if (mtu > MLX4_EN_MAX_XDP_MTU) {
2279                 en_err(priv, "mtu:%d > max:%d when XDP prog is attached\n",
2280                        mtu, MLX4_EN_MAX_XDP_MTU);
2281                 return false;
2282         }
2283
2284         return true;
2285 }
2286
2287 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
2288 {
2289         struct mlx4_en_priv *priv = netdev_priv(dev);
2290         struct mlx4_en_dev *mdev = priv->mdev;
2291         int err = 0;
2292
2293         en_dbg(DRV, priv, "Change MTU called - current:%d new:%d\n",
2294                  dev->mtu, new_mtu);
2295
2296         if (priv->tx_ring_num[TX_XDP] &&
2297             !mlx4_en_check_xdp_mtu(dev, new_mtu))
2298                 return -EOPNOTSUPP;
2299
2300         dev->mtu = new_mtu;
2301
2302         if (netif_running(dev)) {
2303                 mutex_lock(&mdev->state_lock);
2304                 if (!mdev->device_up) {
2305                         /* NIC is probably restarting - let watchdog task reset
2306                          * the port */
2307                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
2308                 } else {
2309                         mlx4_en_stop_port(dev, 1);
2310                         err = mlx4_en_start_port(dev);
2311                         if (err) {
2312                                 en_err(priv, "Failed restarting port:%d\n",
2313                                          priv->port);
2314                                 queue_work(mdev->workqueue, &priv->watchdog_task);
2315                         }
2316                 }
2317                 mutex_unlock(&mdev->state_lock);
2318         }
2319         return 0;
2320 }
2321
2322 static int mlx4_en_hwtstamp_set(struct net_device *dev, struct ifreq *ifr)
2323 {
2324         struct mlx4_en_priv *priv = netdev_priv(dev);
2325         struct mlx4_en_dev *mdev = priv->mdev;
2326         struct hwtstamp_config config;
2327
2328         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
2329                 return -EFAULT;
2330
2331         /* reserved for future extensions */
2332         if (config.flags)
2333                 return -EINVAL;
2334
2335         /* device doesn't support time stamping */
2336         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS))
2337                 return -EINVAL;
2338
2339         /* TX HW timestamp */
2340         switch (config.tx_type) {
2341         case HWTSTAMP_TX_OFF:
2342         case HWTSTAMP_TX_ON:
2343                 break;
2344         default:
2345                 return -ERANGE;
2346         }
2347
2348         /* RX HW timestamp */
2349         switch (config.rx_filter) {
2350         case HWTSTAMP_FILTER_NONE:
2351                 break;
2352         case HWTSTAMP_FILTER_ALL:
2353         case HWTSTAMP_FILTER_SOME:
2354         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2355         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2356         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2357         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2358         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2359         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2360         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
2361         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
2362         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
2363         case HWTSTAMP_FILTER_PTP_V2_EVENT:
2364         case HWTSTAMP_FILTER_PTP_V2_SYNC:
2365         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
2366                 config.rx_filter = HWTSTAMP_FILTER_ALL;
2367                 break;
2368         default:
2369                 return -ERANGE;
2370         }
2371
2372         if (mlx4_en_reset_config(dev, config, dev->features)) {
2373                 config.tx_type = HWTSTAMP_TX_OFF;
2374                 config.rx_filter = HWTSTAMP_FILTER_NONE;
2375         }
2376
2377         return copy_to_user(ifr->ifr_data, &config,
2378                             sizeof(config)) ? -EFAULT : 0;
2379 }
2380
2381 static int mlx4_en_hwtstamp_get(struct net_device *dev, struct ifreq *ifr)
2382 {
2383         struct mlx4_en_priv *priv = netdev_priv(dev);
2384
2385         return copy_to_user(ifr->ifr_data, &priv->hwtstamp_config,
2386                             sizeof(priv->hwtstamp_config)) ? -EFAULT : 0;
2387 }
2388
2389 static int mlx4_en_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
2390 {
2391         switch (cmd) {
2392         case SIOCSHWTSTAMP:
2393                 return mlx4_en_hwtstamp_set(dev, ifr);
2394         case SIOCGHWTSTAMP:
2395                 return mlx4_en_hwtstamp_get(dev, ifr);
2396         default:
2397                 return -EOPNOTSUPP;
2398         }
2399 }
2400
2401 static netdev_features_t mlx4_en_fix_features(struct net_device *netdev,
2402                                               netdev_features_t features)
2403 {
2404         struct mlx4_en_priv *en_priv = netdev_priv(netdev);
2405         struct mlx4_en_dev *mdev = en_priv->mdev;
2406
2407         /* Since there is no support for separate RX C-TAG/S-TAG vlan accel
2408          * enable/disable make sure S-TAG flag is always in same state as
2409          * C-TAG.
2410          */
2411         if (features & NETIF_F_HW_VLAN_CTAG_RX &&
2412             !(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
2413                 features |= NETIF_F_HW_VLAN_STAG_RX;
2414         else
2415                 features &= ~NETIF_F_HW_VLAN_STAG_RX;
2416
2417         return features;
2418 }
2419
2420 static int mlx4_en_set_features(struct net_device *netdev,
2421                 netdev_features_t features)
2422 {
2423         struct mlx4_en_priv *priv = netdev_priv(netdev);
2424         bool reset = false;
2425         int ret = 0;
2426
2427         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXFCS)) {
2428                 en_info(priv, "Turn %s RX-FCS\n",
2429                         (features & NETIF_F_RXFCS) ? "ON" : "OFF");
2430                 reset = true;
2431         }
2432
2433         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_RXALL)) {
2434                 u8 ignore_fcs_value = (features & NETIF_F_RXALL) ? 1 : 0;
2435
2436                 en_info(priv, "Turn %s RX-ALL\n",
2437                         ignore_fcs_value ? "ON" : "OFF");
2438                 ret = mlx4_SET_PORT_fcs_check(priv->mdev->dev,
2439                                               priv->port, ignore_fcs_value);
2440                 if (ret)
2441                         return ret;
2442         }
2443
2444         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
2445                 en_info(priv, "Turn %s RX vlan strip offload\n",
2446                         (features & NETIF_F_HW_VLAN_CTAG_RX) ? "ON" : "OFF");
2447                 reset = true;
2448         }
2449
2450         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_CTAG_TX))
2451                 en_info(priv, "Turn %s TX vlan strip offload\n",
2452                         (features & NETIF_F_HW_VLAN_CTAG_TX) ? "ON" : "OFF");
2453
2454         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_HW_VLAN_STAG_TX))
2455                 en_info(priv, "Turn %s TX S-VLAN strip offload\n",
2456                         (features & NETIF_F_HW_VLAN_STAG_TX) ? "ON" : "OFF");
2457
2458         if (DEV_FEATURE_CHANGED(netdev, features, NETIF_F_LOOPBACK)) {
2459                 en_info(priv, "Turn %s loopback\n",
2460                         (features & NETIF_F_LOOPBACK) ? "ON" : "OFF");
2461                 mlx4_en_update_loopback_state(netdev, features);
2462         }
2463
2464         if (reset) {
2465                 ret = mlx4_en_reset_config(netdev, priv->hwtstamp_config,
2466                                            features);
2467                 if (ret)
2468                         return ret;
2469         }
2470
2471         return 0;
2472 }
2473
2474 static int mlx4_en_set_vf_mac(struct net_device *dev, int queue, u8 *mac)
2475 {
2476         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2477         struct mlx4_en_dev *mdev = en_priv->mdev;
2478         u64 mac_u64 = mlx4_mac_to_u64(mac);
2479
2480         if (is_multicast_ether_addr(mac))
2481                 return -EINVAL;
2482
2483         return mlx4_set_vf_mac(mdev->dev, en_priv->port, queue, mac_u64);
2484 }
2485
2486 static int mlx4_en_set_vf_vlan(struct net_device *dev, int vf, u16 vlan, u8 qos,
2487                                __be16 vlan_proto)
2488 {
2489         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2490         struct mlx4_en_dev *mdev = en_priv->mdev;
2491
2492         return mlx4_set_vf_vlan(mdev->dev, en_priv->port, vf, vlan, qos,
2493                                 vlan_proto);
2494 }
2495
2496 static int mlx4_en_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2497                                int max_tx_rate)
2498 {
2499         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2500         struct mlx4_en_dev *mdev = en_priv->mdev;
2501
2502         return mlx4_set_vf_rate(mdev->dev, en_priv->port, vf, min_tx_rate,
2503                                 max_tx_rate);
2504 }
2505
2506 static int mlx4_en_set_vf_spoofchk(struct net_device *dev, int vf, bool setting)
2507 {
2508         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2509         struct mlx4_en_dev *mdev = en_priv->mdev;
2510
2511         return mlx4_set_vf_spoofchk(mdev->dev, en_priv->port, vf, setting);
2512 }
2513
2514 static int mlx4_en_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivf)
2515 {
2516         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2517         struct mlx4_en_dev *mdev = en_priv->mdev;
2518
2519         return mlx4_get_vf_config(mdev->dev, en_priv->port, vf, ivf);
2520 }
2521
2522 static int mlx4_en_set_vf_link_state(struct net_device *dev, int vf, int link_state)
2523 {
2524         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2525         struct mlx4_en_dev *mdev = en_priv->mdev;
2526
2527         return mlx4_set_vf_link_state(mdev->dev, en_priv->port, vf, link_state);
2528 }
2529
2530 static int mlx4_en_get_vf_stats(struct net_device *dev, int vf,
2531                                 struct ifla_vf_stats *vf_stats)
2532 {
2533         struct mlx4_en_priv *en_priv = netdev_priv(dev);
2534         struct mlx4_en_dev *mdev = en_priv->mdev;
2535
2536         return mlx4_get_vf_stats(mdev->dev, en_priv->port, vf, vf_stats);
2537 }
2538
2539 #define PORT_ID_BYTE_LEN 8
2540 static int mlx4_en_get_phys_port_id(struct net_device *dev,
2541                                     struct netdev_phys_item_id *ppid)
2542 {
2543         struct mlx4_en_priv *priv = netdev_priv(dev);
2544         struct mlx4_dev *mdev = priv->mdev->dev;
2545         int i;
2546         u64 phys_port_id = mdev->caps.phys_port_id[priv->port];
2547
2548         if (!phys_port_id)
2549                 return -EOPNOTSUPP;
2550
2551         ppid->id_len = sizeof(phys_port_id);
2552         for (i = PORT_ID_BYTE_LEN - 1; i >= 0; --i) {
2553                 ppid->id[i] =  phys_port_id & 0xff;
2554                 phys_port_id >>= 8;
2555         }
2556         return 0;
2557 }
2558
2559 static void mlx4_en_add_vxlan_offloads(struct work_struct *work)
2560 {
2561         int ret;
2562         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2563                                                  vxlan_add_task);
2564
2565         ret = mlx4_config_vxlan_port(priv->mdev->dev, priv->vxlan_port);
2566         if (ret)
2567                 goto out;
2568
2569         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2570                                   VXLAN_STEER_BY_OUTER_MAC, 1);
2571 out:
2572         if (ret) {
2573                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2574                 return;
2575         }
2576
2577         /* set offloads */
2578         priv->dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2579                                       NETIF_F_RXCSUM |
2580                                       NETIF_F_TSO | NETIF_F_TSO6 |
2581                                       NETIF_F_GSO_UDP_TUNNEL |
2582                                       NETIF_F_GSO_UDP_TUNNEL_CSUM |
2583                                       NETIF_F_GSO_PARTIAL;
2584 }
2585
2586 static void mlx4_en_del_vxlan_offloads(struct work_struct *work)
2587 {
2588         int ret;
2589         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
2590                                                  vxlan_del_task);
2591         /* unset offloads */
2592         priv->dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
2593                                         NETIF_F_RXCSUM |
2594                                         NETIF_F_TSO | NETIF_F_TSO6 |
2595                                         NETIF_F_GSO_UDP_TUNNEL |
2596                                         NETIF_F_GSO_UDP_TUNNEL_CSUM |
2597                                         NETIF_F_GSO_PARTIAL);
2598
2599         ret = mlx4_SET_PORT_VXLAN(priv->mdev->dev, priv->port,
2600                                   VXLAN_STEER_BY_OUTER_MAC, 0);
2601         if (ret)
2602                 en_err(priv, "failed setting L2 tunnel configuration ret %d\n", ret);
2603
2604         priv->vxlan_port = 0;
2605 }
2606
2607 static void mlx4_en_add_vxlan_port(struct  net_device *dev,
2608                                    struct udp_tunnel_info *ti)
2609 {
2610         struct mlx4_en_priv *priv = netdev_priv(dev);
2611         __be16 port = ti->port;
2612         __be16 current_port;
2613
2614         if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2615                 return;
2616
2617         if (ti->sa_family != AF_INET)
2618                 return;
2619
2620         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2621                 return;
2622
2623         current_port = priv->vxlan_port;
2624         if (current_port && current_port != port) {
2625                 en_warn(priv, "vxlan port %d configured, can't add port %d\n",
2626                         ntohs(current_port), ntohs(port));
2627                 return;
2628         }
2629
2630         priv->vxlan_port = port;
2631         queue_work(priv->mdev->workqueue, &priv->vxlan_add_task);
2632 }
2633
2634 static void mlx4_en_del_vxlan_port(struct  net_device *dev,
2635                                    struct udp_tunnel_info *ti)
2636 {
2637         struct mlx4_en_priv *priv = netdev_priv(dev);
2638         __be16 port = ti->port;
2639         __be16 current_port;
2640
2641         if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2642                 return;
2643
2644         if (ti->sa_family != AF_INET)
2645                 return;
2646
2647         if (priv->mdev->dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
2648                 return;
2649
2650         current_port = priv->vxlan_port;
2651         if (current_port != port) {
2652                 en_dbg(DRV, priv, "vxlan port %d isn't configured, ignoring\n", ntohs(port));
2653                 return;
2654         }
2655
2656         queue_work(priv->mdev->workqueue, &priv->vxlan_del_task);
2657 }
2658
2659 static netdev_features_t mlx4_en_features_check(struct sk_buff *skb,
2660                                                 struct net_device *dev,
2661                                                 netdev_features_t features)
2662 {
2663         features = vlan_features_check(skb, features);
2664         features = vxlan_features_check(skb, features);
2665
2666         /* The ConnectX-3 doesn't support outer IPv6 checksums but it does
2667          * support inner IPv6 checksums and segmentation so  we need to
2668          * strip that feature if this is an IPv6 encapsulated frame.
2669          */
2670         if (skb->encapsulation &&
2671             (skb->ip_summed == CHECKSUM_PARTIAL)) {
2672                 struct mlx4_en_priv *priv = netdev_priv(dev);
2673
2674                 if (!priv->vxlan_port ||
2675                     (ip_hdr(skb)->version != 4) ||
2676                     (udp_hdr(skb)->dest != priv->vxlan_port))
2677                         features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2678         }
2679
2680         return features;
2681 }
2682
2683 static int mlx4_en_set_tx_maxrate(struct net_device *dev, int queue_index, u32 maxrate)
2684 {
2685         struct mlx4_en_priv *priv = netdev_priv(dev);
2686         struct mlx4_en_tx_ring *tx_ring = priv->tx_ring[TX][queue_index];
2687         struct mlx4_update_qp_params params;
2688         int err;
2689
2690         if (!(priv->mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QP_RATE_LIMIT))
2691                 return -EOPNOTSUPP;
2692
2693         /* rate provided to us in Mbs, check if it fits into 12 bits, if not use Gbs */
2694         if (maxrate >> 12) {
2695                 params.rate_unit = MLX4_QP_RATE_LIMIT_GBS;
2696                 params.rate_val  = maxrate / 1000;
2697         } else if (maxrate) {
2698                 params.rate_unit = MLX4_QP_RATE_LIMIT_MBS;
2699                 params.rate_val  = maxrate;
2700         } else { /* zero serves to revoke the QP rate-limitation */
2701                 params.rate_unit = 0;
2702                 params.rate_val  = 0;
2703         }
2704
2705         err = mlx4_update_qp(priv->mdev->dev, tx_ring->qpn, MLX4_UPDATE_QP_RATE_LIMIT,
2706                              &params);
2707         return err;
2708 }
2709
2710 static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
2711 {
2712         struct mlx4_en_priv *priv = netdev_priv(dev);
2713         struct mlx4_en_dev *mdev = priv->mdev;
2714         struct mlx4_en_port_profile new_prof;
2715         struct bpf_prog *old_prog;
2716         struct mlx4_en_priv *tmp;
2717         int tx_changed = 0;
2718         int xdp_ring_num;
2719         int port_up = 0;
2720         int err;
2721         int i;
2722
2723         xdp_ring_num = prog ? priv->rx_ring_num : 0;
2724
2725         /* No need to reconfigure buffers when simply swapping the
2726          * program for a new one.
2727          */
2728         if (priv->tx_ring_num[TX_XDP] == xdp_ring_num) {
2729                 if (prog) {
2730                         prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2731                         if (IS_ERR(prog))
2732                                 return PTR_ERR(prog);
2733                 }
2734                 mutex_lock(&mdev->state_lock);
2735                 for (i = 0; i < priv->rx_ring_num; i++) {
2736                         old_prog = rcu_dereference_protected(
2737                                         priv->rx_ring[i]->xdp_prog,
2738                                         lockdep_is_held(&mdev->state_lock));
2739                         rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2740                         if (old_prog)
2741                                 bpf_prog_put(old_prog);
2742                 }
2743                 mutex_unlock(&mdev->state_lock);
2744                 return 0;
2745         }
2746
2747         if (!mlx4_en_check_xdp_mtu(dev, dev->mtu))
2748                 return -EOPNOTSUPP;
2749
2750         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
2751         if (!tmp)
2752                 return -ENOMEM;
2753
2754         if (prog) {
2755                 prog = bpf_prog_add(prog, priv->rx_ring_num - 1);
2756                 if (IS_ERR(prog)) {
2757                         err = PTR_ERR(prog);
2758                         goto out;
2759                 }
2760         }
2761
2762         mutex_lock(&mdev->state_lock);
2763         memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
2764         new_prof.tx_ring_num[TX_XDP] = xdp_ring_num;
2765
2766         if (priv->tx_ring_num[TX] + xdp_ring_num > MAX_TX_RINGS) {
2767                 tx_changed = 1;
2768                 new_prof.tx_ring_num[TX] =
2769                         MAX_TX_RINGS - ALIGN(xdp_ring_num, MLX4_EN_NUM_UP);
2770                 en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
2771         }
2772
2773         err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
2774         if (err) {
2775                 if (prog)
2776                         bpf_prog_sub(prog, priv->rx_ring_num - 1);
2777                 goto unlock_out;
2778         }
2779
2780         if (priv->port_up) {
2781                 port_up = 1;
2782                 mlx4_en_stop_port(dev, 1);
2783         }
2784
2785         mlx4_en_safe_replace_resources(priv, tmp);
2786         if (tx_changed)
2787                 netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
2788
2789         for (i = 0; i < priv->rx_ring_num; i++) {
2790                 old_prog = rcu_dereference_protected(
2791                                         priv->rx_ring[i]->xdp_prog,
2792                                         lockdep_is_held(&mdev->state_lock));
2793                 rcu_assign_pointer(priv->rx_ring[i]->xdp_prog, prog);
2794                 if (old_prog)
2795                         bpf_prog_put(old_prog);
2796         }
2797
2798         if (port_up) {
2799                 err = mlx4_en_start_port(dev);
2800                 if (err) {
2801                         en_err(priv, "Failed starting port %d for XDP change\n",
2802                                priv->port);
2803                         queue_work(mdev->workqueue, &priv->watchdog_task);
2804                 }
2805         }
2806
2807 unlock_out:
2808         mutex_unlock(&mdev->state_lock);
2809 out:
2810         kfree(tmp);
2811         return err;
2812 }
2813
2814 static bool mlx4_xdp_attached(struct net_device *dev)
2815 {
2816         struct mlx4_en_priv *priv = netdev_priv(dev);
2817
2818         return !!priv->tx_ring_num[TX_XDP];
2819 }
2820
2821 static int mlx4_xdp(struct net_device *dev, struct netdev_xdp *xdp)
2822 {
2823         switch (xdp->command) {
2824         case XDP_SETUP_PROG:
2825                 return mlx4_xdp_set(dev, xdp->prog);
2826         case XDP_QUERY_PROG:
2827                 xdp->prog_attached = mlx4_xdp_attached(dev);
2828                 return 0;
2829         default:
2830                 return -EINVAL;
2831         }
2832 }
2833
2834 static const struct net_device_ops mlx4_netdev_ops = {
2835         .ndo_open               = mlx4_en_open,
2836         .ndo_stop               = mlx4_en_close,
2837         .ndo_start_xmit         = mlx4_en_xmit,
2838         .ndo_select_queue       = mlx4_en_select_queue,
2839         .ndo_get_stats64        = mlx4_en_get_stats64,
2840         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2841         .ndo_set_mac_address    = mlx4_en_set_mac,
2842         .ndo_validate_addr      = eth_validate_addr,
2843         .ndo_change_mtu         = mlx4_en_change_mtu,
2844         .ndo_do_ioctl           = mlx4_en_ioctl,
2845         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2846         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2847         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2848 #ifdef CONFIG_NET_POLL_CONTROLLER
2849         .ndo_poll_controller    = mlx4_en_netpoll,
2850 #endif
2851         .ndo_set_features       = mlx4_en_set_features,
2852         .ndo_fix_features       = mlx4_en_fix_features,
2853         .ndo_setup_tc           = __mlx4_en_setup_tc,
2854 #ifdef CONFIG_RFS_ACCEL
2855         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2856 #endif
2857         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2858         .ndo_udp_tunnel_add     = mlx4_en_add_vxlan_port,
2859         .ndo_udp_tunnel_del     = mlx4_en_del_vxlan_port,
2860         .ndo_features_check     = mlx4_en_features_check,
2861         .ndo_set_tx_maxrate     = mlx4_en_set_tx_maxrate,
2862         .ndo_xdp                = mlx4_xdp,
2863 };
2864
2865 static const struct net_device_ops mlx4_netdev_ops_master = {
2866         .ndo_open               = mlx4_en_open,
2867         .ndo_stop               = mlx4_en_close,
2868         .ndo_start_xmit         = mlx4_en_xmit,
2869         .ndo_select_queue       = mlx4_en_select_queue,
2870         .ndo_get_stats64        = mlx4_en_get_stats64,
2871         .ndo_set_rx_mode        = mlx4_en_set_rx_mode,
2872         .ndo_set_mac_address    = mlx4_en_set_mac,
2873         .ndo_validate_addr      = eth_validate_addr,
2874         .ndo_change_mtu         = mlx4_en_change_mtu,
2875         .ndo_tx_timeout         = mlx4_en_tx_timeout,
2876         .ndo_vlan_rx_add_vid    = mlx4_en_vlan_rx_add_vid,
2877         .ndo_vlan_rx_kill_vid   = mlx4_en_vlan_rx_kill_vid,
2878         .ndo_set_vf_mac         = mlx4_en_set_vf_mac,
2879         .ndo_set_vf_vlan        = mlx4_en_set_vf_vlan,
2880         .ndo_set_vf_rate        = mlx4_en_set_vf_rate,
2881         .ndo_set_vf_spoofchk    = mlx4_en_set_vf_spoofchk,
2882         .ndo_set_vf_link_state  = mlx4_en_set_vf_link_state,
2883         .ndo_get_vf_stats       = mlx4_en_get_vf_stats,
2884         .ndo_get_vf_config      = mlx4_en_get_vf_config,
2885 #ifdef CONFIG_NET_POLL_CONTROLLER
2886         .ndo_poll_controller    = mlx4_en_netpoll,
2887 #endif
2888         .ndo_set_features       = mlx4_en_set_features,
2889         .ndo_fix_features       = mlx4_en_fix_features,
2890         .ndo_setup_tc           = __mlx4_en_setup_tc,
2891 #ifdef CONFIG_RFS_ACCEL
2892         .ndo_rx_flow_steer      = mlx4_en_filter_rfs,
2893 #endif
2894         .ndo_get_phys_port_id   = mlx4_en_get_phys_port_id,
2895         .ndo_udp_tunnel_add     = mlx4_en_add_vxlan_port,
2896         .ndo_udp_tunnel_del     = mlx4_en_del_vxlan_port,
2897         .ndo_features_check     = mlx4_en_features_check,
2898         .ndo_set_tx_maxrate     = mlx4_en_set_tx_maxrate,
2899         .ndo_xdp                = mlx4_xdp,
2900 };
2901
2902 struct mlx4_en_bond {
2903         struct work_struct work;
2904         struct mlx4_en_priv *priv;
2905         int is_bonded;
2906         struct mlx4_port_map port_map;
2907 };
2908
2909 static void mlx4_en_bond_work(struct work_struct *work)
2910 {
2911         struct mlx4_en_bond *bond = container_of(work,
2912                                                      struct mlx4_en_bond,
2913                                                      work);
2914         int err = 0;
2915         struct mlx4_dev *dev = bond->priv->mdev->dev;
2916
2917         if (bond->is_bonded) {
2918                 if (!mlx4_is_bonded(dev)) {
2919                         err = mlx4_bond(dev);
2920                         if (err)
2921                                 en_err(bond->priv, "Fail to bond device\n");
2922                 }
2923                 if (!err) {
2924                         err = mlx4_port_map_set(dev, &bond->port_map);
2925                         if (err)
2926                                 en_err(bond->priv, "Fail to set port map [%d][%d]: %d\n",
2927                                        bond->port_map.port1,
2928                                        bond->port_map.port2,
2929                                        err);
2930                 }
2931         } else if (mlx4_is_bonded(dev)) {
2932                 err = mlx4_unbond(dev);
2933                 if (err)
2934                         en_err(bond->priv, "Fail to unbond device\n");
2935         }
2936         dev_put(bond->priv->dev);
2937         kfree(bond);
2938 }
2939
2940 static int mlx4_en_queue_bond_work(struct mlx4_en_priv *priv, int is_bonded,
2941                                    u8 v2p_p1, u8 v2p_p2)
2942 {
2943         struct mlx4_en_bond *bond = NULL;
2944
2945         bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
2946         if (!bond)
2947                 return -ENOMEM;
2948
2949         INIT_WORK(&bond->work, mlx4_en_bond_work);
2950         bond->priv = priv;
2951         bond->is_bonded = is_bonded;
2952         bond->port_map.port1 = v2p_p1;
2953         bond->port_map.port2 = v2p_p2;
2954         dev_hold(priv->dev);
2955         queue_work(priv->mdev->workqueue, &bond->work);
2956         return 0;
2957 }
2958
2959 int mlx4_en_netdev_event(struct notifier_block *this,
2960                          unsigned long event, void *ptr)
2961 {
2962         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
2963         u8 port = 0;
2964         struct mlx4_en_dev *mdev;
2965         struct mlx4_dev *dev;
2966         int i, num_eth_ports = 0;
2967         bool do_bond = true;
2968         struct mlx4_en_priv *priv;
2969         u8 v2p_port1 = 0;
2970         u8 v2p_port2 = 0;
2971
2972         if (!net_eq(dev_net(ndev), &init_net))
2973                 return NOTIFY_DONE;
2974
2975         mdev = container_of(this, struct mlx4_en_dev, nb);
2976         dev = mdev->dev;
2977
2978         /* Go into this mode only when two network devices set on two ports
2979          * of the same mlx4 device are slaves of the same bonding master
2980          */
2981         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
2982                 ++num_eth_ports;
2983                 if (!port && (mdev->pndev[i] == ndev))
2984                         port = i;
2985                 mdev->upper[i] = mdev->pndev[i] ?
2986                         netdev_master_upper_dev_get(mdev->pndev[i]) : NULL;
2987                 /* condition not met: network device is a slave */
2988                 if (!mdev->upper[i])
2989                         do_bond = false;
2990                 if (num_eth_ports < 2)
2991                         continue;
2992                 /* condition not met: same master */
2993                 if (mdev->upper[i] != mdev->upper[i-1])
2994                         do_bond = false;
2995         }
2996         /* condition not met: 2 salves */
2997         do_bond = (num_eth_ports ==  2) ? do_bond : false;
2998
2999         /* handle only events that come with enough info */
3000         if ((do_bond && (event != NETDEV_BONDING_INFO)) || !port)
3001                 return NOTIFY_DONE;
3002
3003         priv = netdev_priv(ndev);
3004         if (do_bond) {
3005                 struct netdev_notifier_bonding_info *notifier_info = ptr;
3006                 struct netdev_bonding_info *bonding_info =
3007                         &notifier_info->bonding_info;
3008
3009                 /* required mode 1, 2 or 4 */
3010                 if ((bonding_info->master.bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3011                     (bonding_info->master.bond_mode != BOND_MODE_XOR) &&
3012                     (bonding_info->master.bond_mode != BOND_MODE_8023AD))
3013                         do_bond = false;
3014
3015                 /* require exactly 2 slaves */
3016                 if (bonding_info->master.num_slaves != 2)
3017                         do_bond = false;
3018
3019                 /* calc v2p */
3020                 if (do_bond) {
3021                         if (bonding_info->master.bond_mode ==
3022                             BOND_MODE_ACTIVEBACKUP) {
3023                                 /* in active-backup mode virtual ports are
3024                                  * mapped to the physical port of the active
3025                                  * slave */
3026                                 if (bonding_info->slave.state ==
3027                                     BOND_STATE_BACKUP) {
3028                                         if (port == 1) {
3029                                                 v2p_port1 = 2;
3030                                                 v2p_port2 = 2;
3031                                         } else {
3032                                                 v2p_port1 = 1;
3033                                                 v2p_port2 = 1;
3034                                         }
3035                                 } else { /* BOND_STATE_ACTIVE */
3036                                         if (port == 1) {
3037                                                 v2p_port1 = 1;
3038                                                 v2p_port2 = 1;
3039                                         } else {
3040                                                 v2p_port1 = 2;
3041                                                 v2p_port2 = 2;
3042                                         }
3043                                 }
3044                         } else { /* Active-Active */
3045                                 /* in active-active mode a virtual port is
3046                                  * mapped to the native physical port if and only
3047                                  * if the physical port is up */
3048                                 __s8 link = bonding_info->slave.link;
3049
3050                                 if (port == 1)
3051                                         v2p_port2 = 2;
3052                                 else
3053                                         v2p_port1 = 1;
3054                                 if ((link == BOND_LINK_UP) ||
3055                                     (link == BOND_LINK_FAIL)) {
3056                                         if (port == 1)
3057                                                 v2p_port1 = 1;
3058                                         else
3059                                                 v2p_port2 = 2;
3060                                 } else { /* BOND_LINK_DOWN || BOND_LINK_BACK */
3061                                         if (port == 1)
3062                                                 v2p_port1 = 2;
3063                                         else
3064                                                 v2p_port2 = 1;
3065                                 }
3066                         }
3067                 }
3068         }
3069
3070         mlx4_en_queue_bond_work(priv, do_bond,
3071                                 v2p_port1, v2p_port2);
3072
3073         return NOTIFY_DONE;
3074 }
3075
3076 void mlx4_en_update_pfc_stats_bitmap(struct mlx4_dev *dev,
3077                                      struct mlx4_en_stats_bitmap *stats_bitmap,
3078                                      u8 rx_ppp, u8 rx_pause,
3079                                      u8 tx_ppp, u8 tx_pause)
3080 {
3081         int last_i = NUM_MAIN_STATS + NUM_PORT_STATS + NUM_PF_STATS;
3082
3083         if (!mlx4_is_slave(dev) &&
3084             (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FLOWSTATS_EN)) {
3085                 mutex_lock(&stats_bitmap->mutex);
3086                 bitmap_clear(stats_bitmap->bitmap, last_i, NUM_FLOW_STATS);
3087
3088                 if (rx_ppp)
3089                         bitmap_set(stats_bitmap->bitmap, last_i,
3090                                    NUM_FLOW_PRIORITY_STATS_RX);
3091                 last_i += NUM_FLOW_PRIORITY_STATS_RX;
3092
3093                 if (rx_pause && !(rx_ppp))
3094                         bitmap_set(stats_bitmap->bitmap, last_i,
3095                                    NUM_FLOW_STATS_RX);
3096                 last_i += NUM_FLOW_STATS_RX;
3097
3098                 if (tx_ppp)
3099                         bitmap_set(stats_bitmap->bitmap, last_i,
3100                                    NUM_FLOW_PRIORITY_STATS_TX);
3101                 last_i += NUM_FLOW_PRIORITY_STATS_TX;
3102
3103                 if (tx_pause && !(tx_ppp))
3104                         bitmap_set(stats_bitmap->bitmap, last_i,
3105                                    NUM_FLOW_STATS_TX);
3106                 last_i += NUM_FLOW_STATS_TX;
3107
3108                 mutex_unlock(&stats_bitmap->mutex);
3109         }
3110 }
3111
3112 void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
3113                               struct mlx4_en_stats_bitmap *stats_bitmap,
3114                               u8 rx_ppp, u8 rx_pause,
3115                               u8 tx_ppp, u8 tx_pause)
3116 {
3117         int last_i = 0;
3118
3119         mutex_init(&stats_bitmap->mutex);
3120         bitmap_zero(stats_bitmap->bitmap, NUM_ALL_STATS);
3121
3122         if (mlx4_is_slave(dev)) {
3123                 bitmap_set(stats_bitmap->bitmap, last_i +
3124                                          MLX4_FIND_NETDEV_STAT(rx_packets), 1);
3125                 bitmap_set(stats_bitmap->bitmap, last_i +
3126                                          MLX4_FIND_NETDEV_STAT(tx_packets), 1);
3127                 bitmap_set(stats_bitmap->bitmap, last_i +
3128                                          MLX4_FIND_NETDEV_STAT(rx_bytes), 1);
3129                 bitmap_set(stats_bitmap->bitmap, last_i +
3130                                          MLX4_FIND_NETDEV_STAT(tx_bytes), 1);
3131                 bitmap_set(stats_bitmap->bitmap, last_i +
3132                                          MLX4_FIND_NETDEV_STAT(rx_dropped), 1);
3133                 bitmap_set(stats_bitmap->bitmap, last_i +
3134                                          MLX4_FIND_NETDEV_STAT(tx_dropped), 1);
3135         } else {
3136                 bitmap_set(stats_bitmap->bitmap, last_i, NUM_MAIN_STATS);
3137         }
3138         last_i += NUM_MAIN_STATS;
3139
3140         bitmap_set(stats_bitmap->bitmap, last_i, NUM_PORT_STATS);
3141         last_i += NUM_PORT_STATS;
3142
3143         if (mlx4_is_master(dev))
3144                 bitmap_set(stats_bitmap->bitmap, last_i,
3145                            NUM_PF_STATS);
3146         last_i += NUM_PF_STATS;
3147
3148         mlx4_en_update_pfc_stats_bitmap(dev, stats_bitmap,
3149                                         rx_ppp, rx_pause,
3150                                         tx_ppp, tx_pause);
3151         last_i += NUM_FLOW_STATS;
3152
3153         if (!mlx4_is_slave(dev))
3154                 bitmap_set(stats_bitmap->bitmap, last_i, NUM_PKT_STATS);
3155         last_i += NUM_PKT_STATS;
3156
3157         bitmap_set(stats_bitmap->bitmap, last_i, NUM_XDP_STATS);
3158         last_i += NUM_XDP_STATS;
3159 }
3160
3161 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
3162                         struct mlx4_en_port_profile *prof)
3163 {
3164         struct net_device *dev;
3165         struct mlx4_en_priv *priv;
3166         int i, t;
3167         int err;
3168
3169         dev = alloc_etherdev_mqs(sizeof(struct mlx4_en_priv),
3170                                  MAX_TX_RINGS, MAX_RX_RINGS);
3171         if (dev == NULL)
3172                 return -ENOMEM;
3173
3174         netif_set_real_num_tx_queues(dev, prof->tx_ring_num[TX]);
3175         netif_set_real_num_rx_queues(dev, prof->rx_ring_num);
3176
3177         SET_NETDEV_DEV(dev, &mdev->dev->persist->pdev->dev);
3178         dev->dev_port = port - 1;
3179
3180         /*
3181          * Initialize driver private data
3182          */
3183
3184         priv = netdev_priv(dev);
3185         memset(priv, 0, sizeof(struct mlx4_en_priv));
3186         priv->counter_index = MLX4_SINK_COUNTER_INDEX(mdev->dev);
3187         spin_lock_init(&priv->stats_lock);
3188         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
3189         INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
3190         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
3191         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
3192         INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
3193         INIT_WORK(&priv->vxlan_add_task, mlx4_en_add_vxlan_offloads);
3194         INIT_WORK(&priv->vxlan_del_task, mlx4_en_del_vxlan_offloads);
3195 #ifdef CONFIG_RFS_ACCEL
3196         INIT_LIST_HEAD(&priv->filters);
3197         spin_lock_init(&priv->filters_lock);
3198 #endif
3199
3200         priv->dev = dev;
3201         priv->mdev = mdev;
3202         priv->ddev = &mdev->pdev->dev;
3203         priv->prof = prof;
3204         priv->port = port;
3205         priv->port_up = false;
3206         priv->flags = prof->flags;
3207         priv->pflags = MLX4_EN_PRIV_FLAGS_BLUEFLAME;
3208         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
3209                         MLX4_WQE_CTRL_SOLICITED);
3210         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
3211         priv->tx_work_limit = MLX4_EN_DEFAULT_TX_WORK;
3212         netdev_rss_key_fill(priv->rss_key, sizeof(priv->rss_key));
3213
3214         for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
3215                 priv->tx_ring_num[t] = prof->tx_ring_num[t];
3216                 if (!priv->tx_ring_num[t])
3217                         continue;
3218
3219                 priv->tx_ring[t] = kzalloc(sizeof(struct mlx4_en_tx_ring *) *
3220                                            MAX_TX_RINGS, GFP_KERNEL);
3221                 if (!priv->tx_ring[t]) {
3222                         err = -ENOMEM;
3223                         goto err_free_tx;
3224                 }
3225                 priv->tx_cq[t] = kzalloc(sizeof(struct mlx4_en_cq *) *
3226                                          MAX_TX_RINGS, GFP_KERNEL);
3227                 if (!priv->tx_cq[t]) {
3228                         kfree(priv->tx_ring[t]);
3229                         err = -ENOMEM;
3230                         goto out;
3231                 }
3232         }
3233         priv->rx_ring_num = prof->rx_ring_num;
3234         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
3235         priv->cqe_size = mdev->dev->caps.cqe_size;
3236         priv->mac_index = -1;
3237         priv->msg_enable = MLX4_EN_MSG_LEVEL;
3238 #ifdef CONFIG_MLX4_EN_DCB
3239         if (!mlx4_is_slave(priv->mdev->dev)) {
3240                 priv->dcbx_cap = DCB_CAP_DCBX_VER_CEE | DCB_CAP_DCBX_HOST |
3241                         DCB_CAP_DCBX_VER_IEEE;
3242                 priv->flags |= MLX4_EN_DCB_ENABLED;
3243                 priv->cee_config.pfc_state = false;
3244
3245                 for (i = 0; i < MLX4_EN_NUM_UP; i++)
3246                         priv->cee_config.dcb_pfc[i] = pfc_disabled;
3247
3248                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
3249                         dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
3250                 } else {
3251                         en_info(priv, "enabling only PFC DCB ops\n");
3252                         dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
3253                 }
3254         }
3255 #endif
3256
3257         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
3258                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
3259
3260         /* Query for default mac and max mtu */
3261         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
3262
3263         if (mdev->dev->caps.rx_checksum_flags_port[priv->port] &
3264             MLX4_RX_CSUM_MODE_VAL_NON_TCP_UDP)
3265                 priv->flags |= MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP;
3266
3267         /* Set default MAC */
3268         dev->addr_len = ETH_ALEN;
3269         mlx4_en_u64_to_mac(dev->dev_addr, mdev->dev->caps.def_mac[priv->port]);
3270         if (!is_valid_ether_addr(dev->dev_addr)) {
3271                 en_err(priv, "Port: %d, invalid mac burned: %pM, quiting\n",
3272                        priv->port, dev->dev_addr);
3273                 err = -EINVAL;
3274                 goto out;
3275         } else if (mlx4_is_slave(priv->mdev->dev) &&
3276                    (priv->mdev->dev->port_random_macs & 1 << priv->port)) {
3277                 /* Random MAC was assigned in mlx4_slave_cap
3278                  * in mlx4_core module
3279                  */
3280                 dev->addr_assign_type |= NET_ADDR_RANDOM;
3281                 en_warn(priv, "Assigned random MAC address %pM\n", dev->dev_addr);
3282         }
3283
3284         memcpy(priv->current_mac, dev->dev_addr, sizeof(priv->current_mac));
3285
3286         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
3287                                           DS_SIZE * MLX4_EN_MAX_RX_FRAGS);
3288         err = mlx4_en_alloc_resources(priv);
3289         if (err)
3290                 goto out;
3291
3292         /* Initialize time stamping config */
3293         priv->hwtstamp_config.flags = 0;
3294         priv->hwtstamp_config.tx_type = HWTSTAMP_TX_OFF;
3295         priv->hwtstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
3296
3297         /* Allocate page for receive rings */
3298         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
3299                                 MLX4_EN_PAGE_SIZE);
3300         if (err) {
3301                 en_err(priv, "Failed to allocate page for rx qps\n");
3302                 goto out;
3303         }
3304         priv->allocated = 1;
3305
3306         /*
3307          * Initialize netdev entry points
3308          */
3309         if (mlx4_is_master(priv->mdev->dev))
3310                 dev->netdev_ops = &mlx4_netdev_ops_master;
3311         else
3312                 dev->netdev_ops = &mlx4_netdev_ops;
3313         dev->watchdog_timeo = MLX4_EN_WATCHDOG_TIMEOUT;
3314         netif_set_real_num_tx_queues(dev, priv->tx_ring_num[TX]);
3315         netif_set_real_num_rx_queues(dev, priv->rx_ring_num);
3316
3317         dev->ethtool_ops = &mlx4_en_ethtool_ops;
3318
3319         /*
3320          * Set driver features
3321          */
3322         dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3323         if (mdev->LSO_support)
3324                 dev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3325
3326         dev->vlan_features = dev->hw_features;
3327
3328         dev->hw_features |= NETIF_F_RXCSUM | NETIF_F_RXHASH;
3329         dev->features = dev->hw_features | NETIF_F_HIGHDMA |
3330                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
3331                         NETIF_F_HW_VLAN_CTAG_FILTER;
3332         dev->hw_features |= NETIF_F_LOOPBACK |
3333                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
3334
3335         if (!(mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN)) {
3336                 dev->features |= NETIF_F_HW_VLAN_STAG_RX |
3337                         NETIF_F_HW_VLAN_STAG_FILTER;
3338                 dev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
3339         }
3340
3341         if (mlx4_is_slave(mdev->dev)) {
3342                 bool vlan_offload_disabled;
3343                 int phv;
3344
3345                 err = get_phv_bit(mdev->dev, port, &phv);
3346                 if (!err && phv) {
3347                         dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3348                         priv->pflags |= MLX4_EN_PRIV_FLAGS_PHV;
3349                 }
3350                 err = mlx4_get_is_vlan_offload_disabled(mdev->dev, port,
3351                                                         &vlan_offload_disabled);
3352                 if (!err && vlan_offload_disabled) {
3353                         dev->hw_features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3354                                               NETIF_F_HW_VLAN_CTAG_RX |
3355                                               NETIF_F_HW_VLAN_STAG_TX |
3356                                               NETIF_F_HW_VLAN_STAG_RX);
3357                         dev->features &= ~(NETIF_F_HW_VLAN_CTAG_TX |
3358                                            NETIF_F_HW_VLAN_CTAG_RX |
3359                                            NETIF_F_HW_VLAN_STAG_TX |
3360                                            NETIF_F_HW_VLAN_STAG_RX);
3361                 }
3362         } else {
3363                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN &&
3364                     !(mdev->dev->caps.flags2 &
3365                       MLX4_DEV_CAP_FLAG2_SKIP_OUTER_VLAN))
3366                         dev->hw_features |= NETIF_F_HW_VLAN_STAG_TX;
3367         }
3368
3369         if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
3370                 dev->hw_features |= NETIF_F_RXFCS;
3371
3372         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS)
3373                 dev->hw_features |= NETIF_F_RXALL;
3374
3375         if (mdev->dev->caps.steering_mode ==
3376             MLX4_STEERING_MODE_DEVICE_MANAGED &&
3377             mdev->dev->caps.dmfs_high_steer_mode != MLX4_STEERING_DMFS_A0_STATIC)
3378                 dev->hw_features |= NETIF_F_NTUPLE;
3379
3380         if (mdev->dev->caps.steering_mode != MLX4_STEERING_MODE_A0)
3381                 dev->priv_flags |= IFF_UNICAST_FLT;
3382
3383         /* Setting a default hash function value */
3384         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP) {
3385                 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3386         } else if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_XOR) {
3387                 priv->rss_hash_fn = ETH_RSS_HASH_XOR;
3388         } else {
3389                 en_warn(priv,
3390                         "No RSS hash capabilities exposed, using Toeplitz\n");
3391                 priv->rss_hash_fn = ETH_RSS_HASH_TOP;
3392         }
3393
3394         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3395                 dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
3396                                     NETIF_F_GSO_UDP_TUNNEL_CSUM |
3397                                     NETIF_F_GSO_PARTIAL;
3398                 dev->features    |= NETIF_F_GSO_UDP_TUNNEL |
3399                                     NETIF_F_GSO_UDP_TUNNEL_CSUM |
3400                                     NETIF_F_GSO_PARTIAL;
3401                 dev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
3402         }
3403
3404         /* MTU range: 46 - hw-specific max */
3405         dev->min_mtu = MLX4_EN_MIN_MTU;
3406         dev->max_mtu = priv->max_mtu;
3407
3408         mdev->pndev[port] = dev;
3409         mdev->upper[port] = NULL;
3410
3411         netif_carrier_off(dev);
3412         mlx4_en_set_default_moderation(priv);
3413
3414         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num[TX]);
3415         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
3416
3417         mlx4_en_update_loopback_state(priv->dev, priv->dev->features);
3418
3419         /* Configure port */
3420         mlx4_en_calc_rx_buf(dev);
3421         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
3422                                     priv->rx_skb_size + ETH_FCS_LEN,
3423                                     prof->tx_pause, prof->tx_ppp,
3424                                     prof->rx_pause, prof->rx_ppp);
3425         if (err) {
3426                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
3427                        priv->port, err);
3428                 goto out;
3429         }
3430
3431         if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
3432                 err = mlx4_SET_PORT_VXLAN(mdev->dev, priv->port, VXLAN_STEER_BY_OUTER_MAC, 1);
3433                 if (err) {
3434                         en_err(priv, "Failed setting port L2 tunnel configuration, err %d\n",
3435                                err);
3436                         goto out;
3437                 }
3438         }
3439
3440         /* Init port */
3441         en_warn(priv, "Initializing port\n");
3442         err = mlx4_INIT_PORT(mdev->dev, priv->port);
3443         if (err) {
3444                 en_err(priv, "Failed Initializing port\n");
3445                 goto out;
3446         }
3447         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
3448
3449         /* Initialize time stamp mechanism */
3450         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
3451                 mlx4_en_init_timestamp(mdev);
3452
3453         queue_delayed_work(mdev->workqueue, &priv->service_task,
3454                            SERVICE_TASK_DELAY);
3455
3456         mlx4_en_set_stats_bitmap(mdev->dev, &priv->stats_bitmap,
3457                                  mdev->profile.prof[priv->port].rx_ppp,
3458                                  mdev->profile.prof[priv->port].rx_pause,
3459                                  mdev->profile.prof[priv->port].tx_ppp,
3460                                  mdev->profile.prof[priv->port].tx_pause);
3461
3462         err = register_netdev(dev);
3463         if (err) {
3464                 en_err(priv, "Netdev registration failed for port %d\n", port);
3465                 goto out;
3466         }
3467
3468         priv->registered = 1;
3469         devlink_port_type_eth_set(mlx4_get_devlink_port(mdev->dev, priv->port),
3470                                   dev);
3471
3472         return 0;
3473
3474 err_free_tx:
3475         while (t--) {
3476                 kfree(priv->tx_ring[t]);
3477                 kfree(priv->tx_cq[t]);
3478         }
3479 out:
3480         mlx4_en_destroy_netdev(dev);
3481         return err;
3482 }
3483
3484 int mlx4_en_reset_config(struct net_device *dev,
3485                          struct hwtstamp_config ts_config,
3486                          netdev_features_t features)
3487 {
3488         struct mlx4_en_priv *priv = netdev_priv(dev);
3489         struct mlx4_en_dev *mdev = priv->mdev;
3490         struct mlx4_en_port_profile new_prof;
3491         struct mlx4_en_priv *tmp;
3492         int port_up = 0;
3493         int err = 0;
3494
3495         if (priv->hwtstamp_config.tx_type == ts_config.tx_type &&
3496             priv->hwtstamp_config.rx_filter == ts_config.rx_filter &&
3497             !DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3498             !DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS))
3499                 return 0; /* Nothing to change */
3500
3501         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX) &&
3502             (features & NETIF_F_HW_VLAN_CTAG_RX) &&
3503             (priv->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE)) {
3504                 en_warn(priv, "Can't turn ON rx vlan offload while time-stamping rx filter is ON\n");
3505                 return -EINVAL;
3506         }
3507
3508         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
3509         if (!tmp)
3510                 return -ENOMEM;
3511
3512         mutex_lock(&mdev->state_lock);
3513
3514         memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
3515         memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
3516
3517         err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
3518         if (err)
3519                 goto out;
3520
3521         if (priv->port_up) {
3522                 port_up = 1;
3523                 mlx4_en_stop_port(dev, 1);
3524         }
3525
3526         en_warn(priv, "Changing device configuration rx filter(%x) rx vlan(%x)\n",
3527                 ts_config.rx_filter,
3528                 !!(features & NETIF_F_HW_VLAN_CTAG_RX));
3529
3530         mlx4_en_safe_replace_resources(priv, tmp);
3531
3532         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_HW_VLAN_CTAG_RX)) {
3533                 if (features & NETIF_F_HW_VLAN_CTAG_RX)
3534                         dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3535                 else
3536                         dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3537         } else if (ts_config.rx_filter == HWTSTAMP_FILTER_NONE) {
3538                 /* RX time-stamping is OFF, update the RX vlan offload
3539                  * to the latest wanted state
3540                  */
3541                 if (dev->wanted_features & NETIF_F_HW_VLAN_CTAG_RX)
3542                         dev->features |= NETIF_F_HW_VLAN_CTAG_RX;
3543                 else
3544                         dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3545         }
3546
3547         if (DEV_FEATURE_CHANGED(dev, features, NETIF_F_RXFCS)) {
3548                 if (features & NETIF_F_RXFCS)
3549                         dev->features |= NETIF_F_RXFCS;
3550                 else
3551                         dev->features &= ~NETIF_F_RXFCS;
3552         }
3553
3554         /* RX vlan offload and RX time-stamping can't co-exist !
3555          * Regardless of the caller's choice,
3556          * Turn Off RX vlan offload in case of time-stamping is ON
3557          */
3558         if (ts_config.rx_filter != HWTSTAMP_FILTER_NONE) {
3559                 if (dev->features & NETIF_F_HW_VLAN_CTAG_RX)
3560                         en_warn(priv, "Turning off RX vlan offload since RX time-stamping is ON\n");
3561                 dev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
3562         }
3563
3564         if (port_up) {
3565                 err = mlx4_en_start_port(dev);
3566                 if (err)
3567                         en_err(priv, "Failed starting port\n");
3568         }
3569
3570 out:
3571         mutex_unlock(&mdev->state_lock);
3572         kfree(tmp);
3573         if (!err)
3574                 netdev_features_change(dev);
3575         return err;
3576 }