sys/ofed/drivers/net/mlx4/en_netdev.c

   1 /*
   2  * Copyright (c) 2007, 2014 Mellanox Technologies. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  *
  32  */
  33
  34 #include <linux/etherdevice.h>
  35 #include <linux/delay.h>
  36 #include <linux/slab.h>
  37 #ifdef CONFIG_NET_RX_BUSY_POLL
  38 #include <net/busy_poll.h>
  39 #endif
  40
  41 #include <linux/list.h>
  42 #include <linux/if_ether.h>
  43
  44 #include <linux/mlx4/driver.h>
  45 #include <linux/mlx4/device.h>
  46 #include <linux/mlx4/cmd.h>
  47 #include <linux/mlx4/cq.h>
  48
  49 #include <sys/sockio.h>
  50 #include <sys/sysctl.h>
  51
  52 #include "mlx4_en.h"
  53 #include "en_port.h"
  54
  55 static void mlx4_en_sysctl_stat(struct mlx4_en_priv *priv);
  56 static void mlx4_en_sysctl_conf(struct mlx4_en_priv *priv);
  57 static int mlx4_en_unit;
  58
  59 #ifdef CONFIG_NET_RX_BUSY_POLL
  60 /* must be called with local_bh_disable()d */
  61 static int mlx4_en_low_latency_recv(struct napi_struct *napi)
  62 {
  63         struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
  64         struct net_device *dev = cq->dev;
  65         struct mlx4_en_priv *priv = netdev_priv(dev);
  66         struct mlx4_en_rx_ring *rx_ring = priv->rx_ring[cq->ring];
  67         int done;
  68
  69         if (!priv->port_up)
  70                 return LL_FLUSH_FAILED;
  71
  72         if (!mlx4_en_cq_lock_poll(cq))
  73                 return LL_FLUSH_BUSY;
  74
  75         done = mlx4_en_process_rx_cq(dev, cq, 4);
  76 #ifdef LL_EXTENDED_STATS
  77         if (done)
  78                 rx_ring->cleaned += done;
  79         else
  80                 rx_ring->misses++;
  81 #endif
  82
  83         mlx4_en_cq_unlock_poll(cq);
  84
  85         return done;
  86 }
  87 #endif  /* CONFIG_NET_RX_BUSY_POLL */
  88
  89 #ifdef CONFIG_RFS_ACCEL
  90
  91 struct mlx4_en_filter {
  92         struct list_head next;
  93         struct work_struct work;
  94
  95         u8     ip_proto;
  96         __be32 src_ip;
  97         __be32 dst_ip;
  98         __be16 src_port;
  99         __be16 dst_port;
 100
 101         int rxq_index;
 102         struct mlx4_en_priv *priv;
 103         u32 flow_id;                    /* RFS infrastructure id */
 104         int id;                         /* mlx4_en driver id */
 105         u64 reg_id;                     /* Flow steering API id */
 106         u8 activated;                   /* Used to prevent expiry before filter
 107                                          * is attached
 108                                          */
 109         struct hlist_node filter_chain;
 110 };
 111
 112 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv);
 113
 114 static enum mlx4_net_trans_rule_id mlx4_ip_proto_to_trans_rule_id(u8 ip_proto)
 115 {
 116         switch (ip_proto) {
 117         case IPPROTO_UDP:
 118                 return MLX4_NET_TRANS_RULE_ID_UDP;
 119         case IPPROTO_TCP:
 120                 return MLX4_NET_TRANS_RULE_ID_TCP;
 121         default:
 122                 return -EPROTONOSUPPORT;
 123         }
 124 };
 125
 126 static void mlx4_en_filter_work(struct work_struct *work)
 127 {
 128         struct mlx4_en_filter *filter = container_of(work,
 129                                                      struct mlx4_en_filter,
 130                                                      work);
 131         struct mlx4_en_priv *priv = filter->priv;
 132         struct mlx4_spec_list spec_tcp_udp = {
 133                 .id = mlx4_ip_proto_to_trans_rule_id(filter->ip_proto),
 134                 {
 135                         .tcp_udp = {
 136                                 .dst_port = filter->dst_port,
 137                                 .dst_port_msk = (__force __be16)-1,
 138                                 .src_port = filter->src_port,
 139                                 .src_port_msk = (__force __be16)-1,
 140                         },
 141                 },
 142         };
 143         struct mlx4_spec_list spec_ip = {
 144                 .id = MLX4_NET_TRANS_RULE_ID_IPV4,
 145                 {
 146                         .ipv4 = {
 147                                 .dst_ip = filter->dst_ip,
 148                                 .dst_ip_msk = (__force __be32)-1,
 149                                 .src_ip = filter->src_ip,
 150                                 .src_ip_msk = (__force __be32)-1,
 151                         },
 152                 },
 153         };
 154         struct mlx4_spec_list spec_eth = {
 155                 .id = MLX4_NET_TRANS_RULE_ID_ETH,
 156         };
 157         struct mlx4_net_trans_rule rule = {
 158                 .list = LIST_HEAD_INIT(rule.list),
 159                 .queue_mode = MLX4_NET_TRANS_Q_LIFO,
 160                 .exclusive = 1,
 161                 .allow_loopback = 1,
 162                 .promisc_mode = MLX4_FS_REGULAR,
 163                 .port = priv->port,
 164                 .priority = MLX4_DOMAIN_RFS,
 165         };
 166         int rc;
 167         __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 168
 169         if (spec_tcp_udp.id < 0) {
 170                 en_warn(priv, "RFS: ignoring unsupported ip protocol (%d)\n",
 171                         filter->ip_proto);
 172                 goto ignore;
 173         }
 174         list_add_tail(&spec_eth.list, &rule.list);
 175         list_add_tail(&spec_ip.list, &rule.list);
 176         list_add_tail(&spec_tcp_udp.list, &rule.list);
 177
 178         rule.qpn = priv->rss_map.qps[filter->rxq_index].qpn;
 179         memcpy(spec_eth.eth.dst_mac, priv->dev->dev_addr, ETH_ALEN);
 180         memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 181
 182         filter->activated = 0;
 183
 184         if (filter->reg_id) {
 185                 rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 186                 if (rc && rc != -ENOENT)
 187                         en_err(priv, "Error detaching flow. rc = %d\n", rc);
 188         }
 189
 190         rc = mlx4_flow_attach(priv->mdev->dev, &rule, &filter->reg_id);
 191         if (rc)
 192                 en_err(priv, "Error attaching flow. err = %d\n", rc);
 193
 194 ignore:
 195         mlx4_en_filter_rfs_expire(priv);
 196
 197         filter->activated = 1;
 198 }
 199
 200 static inline struct hlist_head *
 201 filter_hash_bucket(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 202                    __be16 src_port, __be16 dst_port)
 203 {
 204         unsigned long l;
 205         int bucket_idx;
 206
 207         l = (__force unsigned long)src_port |
 208             ((__force unsigned long)dst_port << 2);
 209         l ^= (__force unsigned long)(src_ip ^ dst_ip);
 210
 211         bucket_idx = hash_long(l, MLX4_EN_FILTER_HASH_SHIFT);
 212
 213         return &priv->filter_hash[bucket_idx];
 214 }
 215
 216 static struct mlx4_en_filter *
 217 mlx4_en_filter_alloc(struct mlx4_en_priv *priv, int rxq_index, __be32 src_ip,
 218                      __be32 dst_ip, u8 ip_proto, __be16 src_port,
 219                      __be16 dst_port, u32 flow_id)
 220 {
 221         struct mlx4_en_filter *filter = NULL;
 222
 223         filter = kzalloc(sizeof(struct mlx4_en_filter), GFP_ATOMIC);
 224         if (!filter)
 225                 return NULL;
 226
 227         filter->priv = priv;
 228         filter->rxq_index = rxq_index;
 229         INIT_WORK(&filter->work, mlx4_en_filter_work);
 230
 231         filter->src_ip = src_ip;
 232         filter->dst_ip = dst_ip;
 233         filter->ip_proto = ip_proto;
 234         filter->src_port = src_port;
 235         filter->dst_port = dst_port;
 236
 237         filter->flow_id = flow_id;
 238
 239         filter->id = priv->last_filter_id++ % RPS_NO_FILTER;
 240
 241         list_add_tail(&filter->next, &priv->filters);
 242         hlist_add_head(&filter->filter_chain,
 243                        filter_hash_bucket(priv, src_ip, dst_ip, src_port,
 244                                           dst_port));
 245
 246         return filter;
 247 }
 248
 249 static void mlx4_en_filter_free(struct mlx4_en_filter *filter)
 250 {
 251         struct mlx4_en_priv *priv = filter->priv;
 252         int rc;
 253
 254         list_del(&filter->next);
 255
 256         rc = mlx4_flow_detach(priv->mdev->dev, filter->reg_id);
 257         if (rc && rc != -ENOENT)
 258                 en_err(priv, "Error detaching flow. rc = %d\n", rc);
 259
 260         kfree(filter);
 261 }
 262
 263 static inline struct mlx4_en_filter *
 264 mlx4_en_filter_find(struct mlx4_en_priv *priv, __be32 src_ip, __be32 dst_ip,
 265                     u8 ip_proto, __be16 src_port, __be16 dst_port)
 266 {
 267         struct hlist_node *elem;
 268         struct mlx4_en_filter *filter;
 269         struct mlx4_en_filter *ret = NULL;
 270
 271         hlist_for_each_entry(filter, elem,
 272                              filter_hash_bucket(priv, src_ip, dst_ip,
 273                                                 src_port, dst_port),
 274                              filter_chain) {
 275                 if (filter->src_ip == src_ip &&
 276                     filter->dst_ip == dst_ip &&
 277                     filter->ip_proto == ip_proto &&
 278                     filter->src_port == src_port &&
 279                     filter->dst_port == dst_port) {
 280                         ret = filter;
 281                         break;
 282                 }
 283         }
 284
 285         return ret;
 286 }
 287
 288 static int
 289 mlx4_en_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
 290                    u16 rxq_index, u32 flow_id)
 291 {
 292         struct mlx4_en_priv *priv = netdev_priv(net_dev);
 293         struct mlx4_en_filter *filter;
 294         const struct iphdr *ip;
 295         const __be16 *ports;
 296         u8 ip_proto;
 297         __be32 src_ip;
 298         __be32 dst_ip;
 299         __be16 src_port;
 300         __be16 dst_port;
 301         int nhoff = skb_network_offset(skb);
 302         int ret = 0;
 303
 304         if (skb->protocol != htons(ETH_P_IP))
 305                 return -EPROTONOSUPPORT;
 306
 307         ip = (const struct iphdr *)(skb->data + nhoff);
 308         if (ip_is_fragment(ip))
 309                 return -EPROTONOSUPPORT;
 310
 311         if ((ip->protocol != IPPROTO_TCP) && (ip->protocol != IPPROTO_UDP))
 312                 return -EPROTONOSUPPORT;
 313         ports = (const __be16 *)(skb->data + nhoff + 4 * ip->ihl);
 314
 315         ip_proto = ip->protocol;
 316         src_ip = ip->saddr;
 317         dst_ip = ip->daddr;
 318         src_port = ports[0];
 319         dst_port = ports[1];
 320
 321         spin_lock_bh(&priv->filters_lock);
 322         filter = mlx4_en_filter_find(priv, src_ip, dst_ip, ip_proto,
 323                                      src_port, dst_port);
 324         if (filter) {
 325                 if (filter->rxq_index == rxq_index)
 326                         goto out;
 327
 328                 filter->rxq_index = rxq_index;
 329         } else {
 330                 filter = mlx4_en_filter_alloc(priv, rxq_index,
 331                                               src_ip, dst_ip, ip_proto,
 332                                               src_port, dst_port, flow_id);
 333                 if (!filter) {
 334                         ret = -ENOMEM;
 335                         goto err;
 336                 }
 337         }
 338
 339         queue_work(priv->mdev->workqueue, &filter->work);
 340
 341 out:
 342         ret = filter->id;
 343 err:
 344         spin_unlock_bh(&priv->filters_lock);
 345
 346         return ret;
 347 }
 348
 349 void mlx4_en_cleanup_filters(struct mlx4_en_priv *priv,
 350                              struct mlx4_en_rx_ring *rx_ring)
 351 {
 352         struct mlx4_en_filter *filter, *tmp;
 353         LIST_HEAD(del_list);
 354
 355         spin_lock_bh(&priv->filters_lock);
 356         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 357                 list_move(&filter->next, &del_list);
 358                 hlist_del(&filter->filter_chain);
 359         }
 360         spin_unlock_bh(&priv->filters_lock);
 361
 362         list_for_each_entry_safe(filter, tmp, &del_list, next) {
 363                 cancel_work_sync(&filter->work);
 364                 mlx4_en_filter_free(filter);
 365         }
 366 }
 367
 368 static void mlx4_en_filter_rfs_expire(struct mlx4_en_priv *priv)
 369 {
 370         struct mlx4_en_filter *filter = NULL, *tmp, *last_filter = NULL;
 371         LIST_HEAD(del_list);
 372         int i = 0;
 373
 374         spin_lock_bh(&priv->filters_lock);
 375         list_for_each_entry_safe(filter, tmp, &priv->filters, next) {
 376                 if (i > MLX4_EN_FILTER_EXPIRY_QUOTA)
 377                         break;
 378
 379                 if (filter->activated &&
 380                     !work_pending(&filter->work) &&
 381                     rps_may_expire_flow(priv->dev,
 382                                         filter->rxq_index, filter->flow_id,
 383                                         filter->id)) {
 384                         list_move(&filter->next, &del_list);
 385                         hlist_del(&filter->filter_chain);
 386                 } else
 387                         last_filter = filter;
 388
 389                 i++;
 390         }
 391
 392         if (last_filter && (&last_filter->next != priv->filters.next))
 393                 list_move(&priv->filters, &last_filter->next);
 394
 395         spin_unlock_bh(&priv->filters_lock);
 396
 397         list_for_each_entry_safe(filter, tmp, &del_list, next)
 398                 mlx4_en_filter_free(filter);
 399 }
 400 #endif
 401
 402 static void mlx4_en_vlan_rx_add_vid(void *arg, struct net_device *dev, u16 vid)
 403 {
 404         struct mlx4_en_priv *priv = netdev_priv(dev);
 405         struct mlx4_en_dev *mdev = priv->mdev;
 406         int err;
 407         int idx;
 408
 409         if (arg != priv)
 410                 return;
 411
 412         en_dbg(HW, priv, "adding VLAN:%d\n", vid);
 413
 414         set_bit(vid, priv->active_vlans);
 415
 416         /* Add VID to port VLAN filter */
 417         mutex_lock(&mdev->state_lock);
 418         if (mdev->device_up && priv->port_up) {
 419                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 420                 if (err)
 421                         en_err(priv, "Failed configuring VLAN filter\n");
 422         }
 423         if (mlx4_register_vlan(mdev->dev, priv->port, vid, &idx))
 424                 en_dbg(HW, priv, "failed adding vlan %d\n", vid);
 425         mutex_unlock(&mdev->state_lock);
 426
 427 }
 428
 429 static void mlx4_en_vlan_rx_kill_vid(void *arg, struct net_device *dev, u16 vid)
 430 {
 431         struct mlx4_en_priv *priv = netdev_priv(dev);
 432         struct mlx4_en_dev *mdev = priv->mdev;
 433         int err;
 434
 435         if (arg != priv)
 436                 return;
 437
 438         en_dbg(HW, priv, "Killing VID:%d\n", vid);
 439
 440         clear_bit(vid, priv->active_vlans);
 441
 442         /* Remove VID from port VLAN filter */
 443         mutex_lock(&mdev->state_lock);
 444         mlx4_unregister_vlan(mdev->dev, priv->port, vid);
 445
 446         if (mdev->device_up && priv->port_up) {
 447                 err = mlx4_SET_VLAN_FLTR(mdev->dev, priv);
 448                 if (err)
 449                         en_err(priv, "Failed configuring VLAN filter\n");
 450         }
 451         mutex_unlock(&mdev->state_lock);
 452
 453 }
 454
 455 static int mlx4_en_uc_steer_add(struct mlx4_en_priv *priv,
 456                                 unsigned char *mac, int *qpn, u64 *reg_id)
 457 {
 458         struct mlx4_en_dev *mdev = priv->mdev;
 459         struct mlx4_dev *dev = mdev->dev;
 460         int err;
 461
 462         switch (dev->caps.steering_mode) {
 463         case MLX4_STEERING_MODE_B0: {
 464                 struct mlx4_qp qp;
 465                 u8 gid[16] = {0};
 466
 467                 qp.qpn = *qpn;
 468                 memcpy(&gid[10], mac, ETH_ALEN);
 469                 gid[5] = priv->port;
 470
 471                 err = mlx4_unicast_attach(dev, &qp, gid, 0, MLX4_PROT_ETH);
 472                 break;
 473         }
 474         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 475                 struct mlx4_spec_list spec_eth = { {NULL} };
 476                 __be64 mac_mask = cpu_to_be64(MLX4_MAC_MASK << 16);
 477
 478                 struct mlx4_net_trans_rule rule = {
 479                         .queue_mode = MLX4_NET_TRANS_Q_FIFO,
 480                         .exclusive = 0,
 481                         .allow_loopback = 1,
 482                         .promisc_mode = MLX4_FS_REGULAR,
 483                         .priority = MLX4_DOMAIN_NIC,
 484                 };
 485
 486                 rule.port = priv->port;
 487                 rule.qpn = *qpn;
 488                 INIT_LIST_HEAD(&rule.list);
 489
 490                 spec_eth.id = MLX4_NET_TRANS_RULE_ID_ETH;
 491                 memcpy(spec_eth.eth.dst_mac, mac, ETH_ALEN);
 492                 memcpy(spec_eth.eth.dst_mac_msk, &mac_mask, ETH_ALEN);
 493                 list_add_tail(&spec_eth.list, &rule.list);
 494
 495                 err = mlx4_flow_attach(dev, &rule, reg_id);
 496                 break;
 497         }
 498         default:
 499                 return -EINVAL;
 500         }
 501         if (err)
 502                 en_warn(priv, "Failed Attaching Unicast\n");
 503
 504         return err;
 505 }
 506
 507 static void mlx4_en_uc_steer_release(struct mlx4_en_priv *priv,
 508                                      unsigned char *mac, int qpn, u64 reg_id)
 509 {
 510         struct mlx4_en_dev *mdev = priv->mdev;
 511         struct mlx4_dev *dev = mdev->dev;
 512
 513         switch (dev->caps.steering_mode) {
 514         case MLX4_STEERING_MODE_B0: {
 515                 struct mlx4_qp qp;
 516                 u8 gid[16] = {0};
 517
 518                 qp.qpn = qpn;
 519                 memcpy(&gid[10], mac, ETH_ALEN);
 520                 gid[5] = priv->port;
 521
 522                 mlx4_unicast_detach(dev, &qp, gid, MLX4_PROT_ETH);
 523                 break;
 524         }
 525         case MLX4_STEERING_MODE_DEVICE_MANAGED: {
 526                 mlx4_flow_detach(dev, reg_id);
 527                 break;
 528         }
 529         default:
 530                 en_err(priv, "Invalid steering mode.\n");
 531         }
 532 }
 533
 534 static int mlx4_en_get_qp(struct mlx4_en_priv *priv)
 535 {
 536         struct mlx4_en_dev *mdev = priv->mdev;
 537         struct mlx4_dev *dev = mdev->dev;
 538         struct mlx4_mac_entry *entry;
 539         int index = 0;
 540         int err = 0;
 541         u64 reg_id;
 542         int *qpn = &priv->base_qpn;
 543         u64 mac = mlx4_mac_to_u64(IF_LLADDR(priv->dev));
 544
 545         en_dbg(DRV, priv, "Registering MAC: %pM for adding\n",
 546                IF_LLADDR(priv->dev));
 547         index = mlx4_register_mac(dev, priv->port, mac);
 548         if (index < 0) {
 549                 err = index;
 550                 en_err(priv, "Failed adding MAC: %pM\n",
 551                        IF_LLADDR(priv->dev));
 552                 return err;
 553         }
 554
 555         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 556                 int base_qpn = mlx4_get_base_qpn(dev, priv->port);
 557                 *qpn = base_qpn + index;
 558                 return 0;
 559         }
 560
 561         err = mlx4_qp_reserve_range(dev, 1, 1, qpn, 0);
 562         en_dbg(DRV, priv, "Reserved qp %d\n", *qpn);
 563         if (err) {
 564                 en_err(priv, "Failed to reserve qp for mac registration\n");
 565                 goto qp_err;
 566         }
 567
 568         err = mlx4_en_uc_steer_add(priv, IF_LLADDR(priv->dev), qpn, &reg_id);
 569         if (err)
 570                 goto steer_err;
 571
 572         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 573         if (!entry) {
 574                 err = -ENOMEM;
 575                 goto alloc_err;
 576         }
 577         memcpy(entry->mac, IF_LLADDR(priv->dev), sizeof(entry->mac));
 578         entry->reg_id = reg_id;
 579
 580         hlist_add_head(&entry->hlist,
 581                            &priv->mac_hash[entry->mac[MLX4_EN_MAC_HASH_IDX]]);
 582
 583         return 0;
 584
 585 alloc_err:
 586         mlx4_en_uc_steer_release(priv, IF_LLADDR(priv->dev), *qpn, reg_id);
 587
 588 steer_err:
 589         mlx4_qp_release_range(dev, *qpn, 1);
 590
 591 qp_err:
 592         mlx4_unregister_mac(dev, priv->port, mac);
 593         return err;
 594 }
 595
 596 static void mlx4_en_put_qp(struct mlx4_en_priv *priv)
 597 {
 598         struct mlx4_en_dev *mdev = priv->mdev;
 599         struct mlx4_dev *dev = mdev->dev;
 600         int qpn = priv->base_qpn;
 601         u64 mac;
 602
 603         if (dev->caps.steering_mode == MLX4_STEERING_MODE_A0) {
 604                 mac = mlx4_mac_to_u64(IF_LLADDR(priv->dev));
 605                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 606                        IF_LLADDR(priv->dev));
 607                 mlx4_unregister_mac(dev, priv->port, mac);
 608         } else {
 609                 struct mlx4_mac_entry *entry;
 610                 struct hlist_node *n, *tmp;
 611                 struct hlist_head *bucket;
 612                 unsigned int i;
 613
 614                 for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i) {
 615                         bucket = &priv->mac_hash[i];
 616                         hlist_for_each_entry_safe(entry, n, tmp, bucket, hlist) {
 617                                 mac = mlx4_mac_to_u64(entry->mac);
 618                                 en_dbg(DRV, priv, "Registering MAC: %pM for deleting\n",
 619                                        entry->mac);
 620                                 mlx4_en_uc_steer_release(priv, entry->mac,
 621                                                          qpn, entry->reg_id);
 622
 623                                 mlx4_unregister_mac(dev, priv->port, mac);
 624                                 hlist_del(&entry->hlist);
 625                                 kfree(entry);
 626                         }
 627                 }
 628
 629                 en_dbg(DRV, priv, "Releasing qp: port %d, qpn %d\n",
 630                        priv->port, qpn);
 631                 mlx4_qp_release_range(dev, qpn, 1);
 632                 priv->flags &= ~MLX4_EN_FLAG_FORCE_PROMISC;
 633         }
 634 }
 635
 636 static void mlx4_en_clear_list(struct net_device *dev)
 637 {
 638         struct mlx4_en_priv *priv = netdev_priv(dev);
 639         struct mlx4_en_mc_list *tmp, *mc_to_del;
 640
 641         list_for_each_entry_safe(mc_to_del, tmp, &priv->mc_list, list) {
 642                 list_del(&mc_to_del->list);
 643                 kfree(mc_to_del);
 644         }
 645 }
 646
 647 static void mlx4_en_cache_mclist(struct net_device *dev)
 648 {
 649         struct ifmultiaddr *ifma;
 650         struct mlx4_en_mc_list *tmp;
 651         struct mlx4_en_priv *priv = netdev_priv(dev);
 652
 653         TAILQ_FOREACH(ifma, &dev->if_multiaddrs, ifma_link) {
 654                 if (ifma->ifma_addr->sa_family != AF_LINK)
 655                         continue;
 656                 if (((struct sockaddr_dl *)ifma->ifma_addr)->sdl_alen !=
 657                                 ETHER_ADDR_LEN)
 658                         continue;
 659                 /* Make sure the list didn't grow. */
 660                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 661                 memcpy(tmp->addr,
 662                         LLADDR((struct sockaddr_dl *)ifma->ifma_addr), ETH_ALEN);
 663                 list_add_tail(&tmp->list, &priv->mc_list);
 664         }
 665 }
 666
 667 static void update_mclist_flags(struct mlx4_en_priv *priv,
 668                                 struct list_head *dst,
 669                                 struct list_head *src)
 670 {
 671         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 672         bool found;
 673
 674         /* Find all the entries that should be removed from dst,
 675          * These are the entries that are not found in src
 676          */
 677         list_for_each_entry(dst_tmp, dst, list) {
 678                 found = false;
 679                 list_for_each_entry(src_tmp, src, list) {
 680                         if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
 681                                 found = true;
 682                                 break;
 683                         }
 684                 }
 685                 if (!found)
 686                         dst_tmp->action = MCLIST_REM;
 687         }
 688
 689         /* Add entries that exist in src but not in dst
 690          * mark them as need to add
 691          */
 692         list_for_each_entry(src_tmp, src, list) {
 693                 found = false;
 694                 list_for_each_entry(dst_tmp, dst, list) {
 695                         if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
 696                                 dst_tmp->action = MCLIST_NONE;
 697                                 found = true;
 698                                 break;
 699                         }
 700                 }
 701                 if (!found) {
 702                         new_mc = kmalloc(sizeof(struct mlx4_en_mc_list),
 703                                          GFP_KERNEL);
 704                         if (!new_mc) {
 705                                 en_err(priv, "Failed to allocate current multicast list\n");
 706                                 return;
 707                         }
 708                         memcpy(new_mc, src_tmp,
 709                                sizeof(struct mlx4_en_mc_list));
 710                         new_mc->action = MCLIST_ADD;
 711                         list_add_tail(&new_mc->list, dst);
 712                 }
 713         }
 714 }
 715
 716 static void mlx4_en_set_rx_mode(struct net_device *dev)
 717 {
 718         struct mlx4_en_priv *priv = netdev_priv(dev);
 719
 720         if (!priv->port_up)
 721                 return;
 722
 723         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 724 }
 725
 726 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 727                                      struct mlx4_en_dev *mdev)
 728 {
 729         int err = 0;
 730         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 731                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 732
 733                 /* Enable promiscouos mode */
 734                 switch (mdev->dev->caps.steering_mode) {
 735                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 736                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 737                                                           priv->port,
 738                                                           priv->base_qpn,
 739                                                           MLX4_FS_ALL_DEFAULT);
 740                         if (err)
 741                                 en_err(priv, "Failed enabling promiscuous mode\n");
 742                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 743                         break;
 744
 745                 case MLX4_STEERING_MODE_B0:
 746                         err = mlx4_unicast_promisc_add(mdev->dev,
 747                                                        priv->base_qpn,
 748                                                        priv->port);
 749                         if (err)
 750                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 751
 752                         /* Add the default qp number as multicast
 753                          * promisc
 754                          */
 755                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 756                                 err = mlx4_multicast_promisc_add(mdev->dev,
 757                                                                  priv->base_qpn,
 758                                                                  priv->port);
 759                                 if (err)
 760                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 761                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 762                         }
 763                         break;
 764
 765                 case MLX4_STEERING_MODE_A0:
 766                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 767                                                      priv->port,
 768                                                      priv->base_qpn,
 769                                                      1);
 770                         if (err)
 771                                 en_err(priv, "Failed enabling promiscuous mode\n");
 772                         break;
 773                 }
 774
 775                 /* Disable port multicast filter (unconditionally) */
 776                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 777                                           0, MLX4_MCAST_DISABLE);
 778                 if (err)
 779                         en_err(priv, "Failed disabling multicast filter\n");
 780         }
 781 }
 782
 783 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 784                                        struct mlx4_en_dev *mdev)
 785 {
 786         int err = 0;
 787
 788         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 789
 790         /* Disable promiscouos mode */
 791         switch (mdev->dev->caps.steering_mode) {
 792         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 793                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 794                                                      priv->port,
 795                                                      MLX4_FS_ALL_DEFAULT);
 796                 if (err)
 797                         en_err(priv, "Failed disabling promiscuous mode\n");
 798                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 799                 break;
 800
 801         case MLX4_STEERING_MODE_B0:
 802                 err = mlx4_unicast_promisc_remove(mdev->dev,
 803                                                   priv->base_qpn,
 804                                                   priv->port);
 805                 if (err)
 806                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 807                 /* Disable Multicast promisc */
 808                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 809                         err = mlx4_multicast_promisc_remove(mdev->dev,
 810                                                             priv->base_qpn,
 811                                                             priv->port);
 812                         if (err)
 813                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 814                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 815                 }
 816                 break;
 817
 818         case MLX4_STEERING_MODE_A0:
 819                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 820                                              priv->port,
 821                                              priv->base_qpn, 0);
 822                 if (err)
 823                         en_err(priv, "Failed disabling promiscuous mode\n");
 824                 break;
 825         }
 826 }
 827
 828 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
 829                                  struct net_device *dev,
 830                                  struct mlx4_en_dev *mdev)
 831 {
 832         struct mlx4_en_mc_list *mclist, *tmp;
 833         u8 mc_list[16] = {0};
 834         int err = 0;
 835         u64 mcast_addr = 0;
 836
 837
 838         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
 839         if (dev->if_flags & IFF_ALLMULTI) {
 840                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 841                                           0, MLX4_MCAST_DISABLE);
 842                 if (err)
 843                         en_err(priv, "Failed disabling multicast filter\n");
 844
 845                 /* Add the default qp number as multicast promisc */
 846                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 847                         switch (mdev->dev->caps.steering_mode) {
 848                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 849                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
 850                                                                   priv->port,
 851                                                                   priv->base_qpn,
 852                                                                   MLX4_FS_MC_DEFAULT);
 853                                 break;
 854
 855                         case MLX4_STEERING_MODE_B0:
 856                                 err = mlx4_multicast_promisc_add(mdev->dev,
 857                                                                  priv->base_qpn,
 858                                                                  priv->port);
 859                                 break;
 860
 861                         case MLX4_STEERING_MODE_A0:
 862                                 break;
 863                         }
 864                         if (err)
 865                                 en_err(priv, "Failed entering multicast promisc mode\n");
 866                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 867                 }
 868         } else {
 869                 /* Disable Multicast promisc */
 870                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 871                         switch (mdev->dev->caps.steering_mode) {
 872                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 873                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 874                                                                      priv->port,
 875                                                                      MLX4_FS_MC_DEFAULT);
 876                                 break;
 877
 878                         case MLX4_STEERING_MODE_B0:
 879                                 err = mlx4_multicast_promisc_remove(mdev->dev,
 880                                                                     priv->base_qpn,
 881                                                                     priv->port);
 882                                 break;
 883
 884                         case MLX4_STEERING_MODE_A0:
 885                                 break;
 886                         }
 887                         if (err)
 888                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 889                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 890                 }
 891
 892                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 893                                           0, MLX4_MCAST_DISABLE);
 894                 if (err)
 895                         en_err(priv, "Failed disabling multicast filter\n");
 896
 897                 /* Flush mcast filter and init it with broadcast address */
 898                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
 899                                     1, MLX4_MCAST_CONFIG);
 900
 901                 /* Update multicast list - we cache all addresses so they won't
 902                  * change while HW is updated holding the command semaphor */
 903                 mlx4_en_cache_mclist(dev);
 904                 list_for_each_entry(mclist, &priv->mc_list, list) {
 905                         mcast_addr = mlx4_mac_to_u64(mclist->addr);
 906                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
 907                                         mcast_addr, 0, MLX4_MCAST_CONFIG);
 908                 }
 909                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 910                                           0, MLX4_MCAST_ENABLE);
 911                 if (err)
 912                         en_err(priv, "Failed enabling multicast filter\n");
 913
 914                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
 915                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
 916                         if (mclist->action == MCLIST_REM) {
 917                                 /* detach this address and delete from list */
 918                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
 919                                 mc_list[5] = priv->port;
 920                                 err = mlx4_multicast_detach(mdev->dev,
 921                                                             &priv->rss_map.indir_qp,
 922                                                             mc_list,
 923                                                             MLX4_PROT_ETH,
 924                                                             mclist->reg_id);
 925                                 if (err)
 926                                         en_err(priv, "Fail to detach multicast address\n");
 927
 928                                 /* remove from list */
 929                                 list_del(&mclist->list);
 930                                 kfree(mclist);
 931                         } else if (mclist->action == MCLIST_ADD) {
 932                                 /* attach the address */
 933                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
 934                                 /* needed for B0 steering support */
 935                                 mc_list[5] = priv->port;
 936                                 err = mlx4_multicast_attach(mdev->dev,
 937                                                             &priv->rss_map.indir_qp,
 938                                                             mc_list,
 939                                                             priv->port, 0,
 940                                                             MLX4_PROT_ETH,
 941                                                             &mclist->reg_id);
 942                                 if (err)
 943                                         en_err(priv, "Fail to attach multicast address\n");
 944
 945                         }
 946                 }
 947         }
 948 }
 949
 950 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
 951 {
 952         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
 953                                                  rx_mode_task);
 954         struct mlx4_en_dev *mdev = priv->mdev;
 955         struct net_device *dev = priv->dev;
 956
 957
 958         mutex_lock(&mdev->state_lock);
 959         if (!mdev->device_up) {
 960                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
 961                 goto out;
 962         }
 963         if (!priv->port_up) {
 964                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
 965                 goto out;
 966         }
 967         if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
 968                 if (priv->port_state.link_state) {
 969                         priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
 970                         /* Important note: the following call for if_link_state_change
 971                          * is needed for interface up scenario (start port, link state
 972                          * change) */
 973                         if_link_state_change(priv->dev, LINK_STATE_UP);
 974                         en_dbg(HW, priv, "Link Up\n");
 975                 }
 976         }
 977
 978         /* Promsicuous mode: disable all filters */
 979         if ((dev->if_flags & IFF_PROMISC) ||
 980             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
 981                 mlx4_en_set_promisc_mode(priv, mdev);
 982                 goto out;
 983         }
 984
 985         /* Not in promiscuous mode */
 986         if (priv->flags & MLX4_EN_FLAG_PROMISC)
 987                 mlx4_en_clear_promisc_mode(priv, mdev);
 988
 989         mlx4_en_do_multicast(priv, dev, mdev);
 990 out:
 991         mutex_unlock(&mdev->state_lock);
 992 }
 993
 994 #ifdef CONFIG_NET_POLL_CONTROLLER
 995 static void mlx4_en_netpoll(struct net_device *dev)
 996 {
 997         struct mlx4_en_priv *priv = netdev_priv(dev);
 998         struct mlx4_en_cq *cq;
 999         unsigned long flags;
1000         int i;
1001
1002         for (i = 0; i < priv->rx_ring_num; i++) {
1003                 cq = priv->rx_cq[i];
1004                 spin_lock_irqsave(&cq->lock, flags);
1005                 napi_synchronize(&cq->napi);
1006                 mlx4_en_process_rx_cq(dev, cq, 0);
1007                 spin_unlock_irqrestore(&cq->lock, flags);
1008         }
1009 }
1010 #endif
1011
1012 static void mlx4_en_watchdog_timeout(void *arg)
1013 {
1014         struct mlx4_en_priv *priv = arg;
1015         struct mlx4_en_dev *mdev = priv->mdev;
1016
1017         en_dbg(DRV, priv, "Scheduling watchdog\n");
1018         queue_work(mdev->workqueue, &priv->watchdog_task);
1019         if (priv->port_up)
1020                 callout_reset(&priv->watchdog_timer, MLX4_EN_WATCHDOG_TIMEOUT,
1021                                 mlx4_en_watchdog_timeout, priv);
1022 }
1023
1024
1025
1026 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1027 {
1028         struct mlx4_en_cq *cq;
1029         int i;
1030
1031         /* If we haven't received a specific coalescing setting
1032          * (module param), we set the moderation parameters as follows:
1033          * - moder_cnt is set to the number of mtu sized packets to
1034          *   satisfy our coelsing target.
1035          * - moder_time is set to a fixed value.
1036          */
1037         priv->rx_frames = MLX4_EN_RX_COAL_TARGET / priv->dev->if_mtu + 1;
1038         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1039         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1040         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1041         en_dbg(INTR, priv, "Default coalesing params for mtu: %u - "
1042                "rx_frames:%d rx_usecs:%d\n",
1043                (unsigned)priv->dev->if_mtu, priv->rx_frames, priv->rx_usecs);
1044
1045         /* Setup cq moderation params */
1046         for (i = 0; i < priv->rx_ring_num; i++) {
1047                 cq = priv->rx_cq[i];
1048                 cq->moder_cnt = priv->rx_frames;
1049                 cq->moder_time = priv->rx_usecs;
1050                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1051                 priv->last_moder_packets[i] = 0;
1052                 priv->last_moder_bytes[i] = 0;
1053         }
1054
1055         for (i = 0; i < priv->tx_ring_num; i++) {
1056                 cq = priv->tx_cq[i];
1057                 cq->moder_cnt = priv->tx_frames;
1058                 cq->moder_time = priv->tx_usecs;
1059         }
1060
1061         /* Reset auto-moderation params */
1062         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1063         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1064         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1065         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1066         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1067         priv->adaptive_rx_coal = 1;
1068         priv->last_moder_jiffies = 0;
1069         priv->last_moder_tx_packets = 0;
1070 }
1071
1072 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1073 {
1074         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1075         struct mlx4_en_cq *cq;
1076         unsigned long packets;
1077         unsigned long rate;
1078         unsigned long avg_pkt_size;
1079         unsigned long rx_packets;
1080         unsigned long rx_bytes;
1081         unsigned long rx_pkt_diff;
1082         int moder_time;
1083         int ring, err;
1084
1085         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1086                 return;
1087
1088         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1089                 spin_lock(&priv->stats_lock);
1090                 rx_packets = priv->rx_ring[ring]->packets;
1091                 rx_bytes = priv->rx_ring[ring]->bytes;
1092                 spin_unlock(&priv->stats_lock);
1093
1094                 rx_pkt_diff = ((unsigned long) (rx_packets -
1095                                 priv->last_moder_packets[ring]));
1096                 packets = rx_pkt_diff;
1097                 rate = packets * HZ / period;
1098                 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1099                                 priv->last_moder_bytes[ring])) / packets : 0;
1100
1101                 /* Apply auto-moderation only when packet rate
1102                  * exceeds a rate that it matters */
1103                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1104                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1105                         if (rate < priv->pkt_rate_low)
1106                                 moder_time = priv->rx_usecs_low;
1107                         else if (rate > priv->pkt_rate_high)
1108                                 moder_time = priv->rx_usecs_high;
1109                         else
1110                                 moder_time = (rate - priv->pkt_rate_low) *
1111                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1112                                         (priv->pkt_rate_high - priv->pkt_rate_low) +
1113                                         priv->rx_usecs_low;
1114                 } else {
1115                         moder_time = priv->rx_usecs_low;
1116                 }
1117
1118                 if (moder_time != priv->last_moder_time[ring]) {
1119                         priv->last_moder_time[ring] = moder_time;
1120                         cq = priv->rx_cq[ring];
1121                         cq->moder_time = moder_time;
1122                         err = mlx4_en_set_cq_moder(priv, cq);
1123                         if (err)
1124                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1125                                        ring);
1126                 }
1127                 priv->last_moder_packets[ring] = rx_packets;
1128                 priv->last_moder_bytes[ring] = rx_bytes;
1129         }
1130
1131         priv->last_moder_jiffies = jiffies;
1132 }
1133
1134 static void mlx4_en_do_get_stats(struct work_struct *work)
1135 {
1136         struct delayed_work *delay = to_delayed_work(work);
1137         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1138                                                  stats_task);
1139         struct mlx4_en_dev *mdev = priv->mdev;
1140         int err;
1141
1142         mutex_lock(&mdev->state_lock);
1143         if (mdev->device_up) {
1144                 if (priv->port_up) {
1145                         err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1146                         if (err)
1147                                 en_dbg(HW, priv, "Could not update stats\n");
1148
1149                         mlx4_en_auto_moderation(priv);
1150                 }
1151
1152                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1153         }
1154         mutex_unlock(&mdev->state_lock);
1155 }
1156
1157 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1158  * periodically
1159  */
1160 static void mlx4_en_service_task(struct work_struct *work)
1161 {
1162         struct delayed_work *delay = to_delayed_work(work);
1163         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1164                                                  service_task);
1165         struct mlx4_en_dev *mdev = priv->mdev;
1166
1167         mutex_lock(&mdev->state_lock);
1168         if (mdev->device_up) {
1169                 queue_delayed_work(mdev->workqueue, &priv->service_task,
1170                                    SERVICE_TASK_DELAY);
1171         }
1172         mutex_unlock(&mdev->state_lock);
1173 }
1174
1175 static void mlx4_en_linkstate(struct work_struct *work)
1176 {
1177         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1178                                                  linkstate_task);
1179         struct mlx4_en_dev *mdev = priv->mdev;
1180         int linkstate = priv->link_state;
1181
1182         mutex_lock(&mdev->state_lock);
1183         /* If observable port state changed set carrier state and
1184          * report to system log */
1185         if (priv->last_link_state != linkstate) {
1186                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1187                         en_info(priv, "Link Down\n");
1188                         if_link_state_change(priv->dev, LINK_STATE_DOWN);
1189                 /* make sure the port is up before notifying the OS.
1190                  * This is tricky since we get here on INIT_PORT and
1191                  * in such case we can't tell the OS the port is up.
1192                  * To solve this there is a call to if_link_state_change
1193                  * in set_rx_mode.
1194                  * */
1195                 } else if (priv->port_up && (linkstate == MLX4_DEV_EVENT_PORT_UP)){
1196                         en_info(priv, "Link Up\n");
1197                         if_link_state_change(priv->dev, LINK_STATE_UP);
1198                 }
1199         }
1200         priv->last_link_state = linkstate;
1201         mutex_unlock(&mdev->state_lock);
1202 }
1203
1204
1205 int mlx4_en_start_port(struct net_device *dev)
1206 {
1207         struct mlx4_en_priv *priv = netdev_priv(dev);
1208         struct mlx4_en_dev *mdev = priv->mdev;
1209         struct mlx4_en_cq *cq;
1210         struct mlx4_en_tx_ring *tx_ring;
1211         int rx_index = 0;
1212         int tx_index = 0;
1213         int err = 0;
1214         int i;
1215         int j;
1216         u8 mc_list[16] = {0};
1217
1218
1219         if (priv->port_up) {
1220                 en_dbg(DRV, priv, "start port called while port already up\n");
1221                 return 0;
1222         }
1223
1224         INIT_LIST_HEAD(&priv->mc_list);
1225         INIT_LIST_HEAD(&priv->curr_list);
1226         INIT_LIST_HEAD(&priv->ethtool_list);
1227
1228         /* Calculate Rx buf size */
1229         dev->if_mtu = min(dev->if_mtu, priv->max_mtu);
1230         mlx4_en_calc_rx_buf(dev);
1231         priv->rx_alloc_size = max_t(int, 2 * roundup_pow_of_two(priv->rx_mb_size),
1232                                     PAGE_SIZE);
1233         priv->rx_alloc_order = get_order(priv->rx_alloc_size);
1234         priv->rx_buf_size = roundup_pow_of_two(priv->rx_mb_size);
1235         priv->log_rx_info = ROUNDUP_LOG2(sizeof(struct mlx4_en_rx_buf));
1236         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_mb_size);
1237
1238         /* Configure rx cq's and rings */
1239         err = mlx4_en_activate_rx_rings(priv);
1240         if (err) {
1241                 en_err(priv, "Failed to activate RX rings\n");
1242                 return err;
1243         }
1244         for (i = 0; i < priv->rx_ring_num; i++) {
1245                 cq = priv->rx_cq[i];
1246
1247                 mlx4_en_cq_init_lock(cq);
1248                 err = mlx4_en_activate_cq(priv, cq, i);
1249                 if (err) {
1250                         en_err(priv, "Failed activating Rx CQ\n");
1251                         goto cq_err;
1252                 }
1253                 for (j = 0; j < cq->size; j++)
1254                         cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1255                 err = mlx4_en_set_cq_moder(priv, cq);
1256                 if (err) {
1257                         en_err(priv, "Failed setting cq moderation parameters");
1258                         mlx4_en_deactivate_cq(priv, cq);
1259                         goto cq_err;
1260                 }
1261                 mlx4_en_arm_cq(priv, cq);
1262                 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1263                 ++rx_index;
1264         }
1265
1266         /* Set qp number */
1267         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1268         err = mlx4_en_get_qp(priv);
1269         if (err) {
1270                 en_err(priv, "Failed getting eth qp\n");
1271                 goto cq_err;
1272         }
1273         mdev->mac_removed[priv->port] = 0;
1274
1275         /* gets default allocated counter index from func cap */
1276         /* or sink counter index if no resources */
1277         priv->counter_index = mdev->dev->caps.def_counter_index[priv->port - 1];
1278
1279         en_dbg(DRV, priv, "%s: default counter index %d for port %d\n",
1280                __func__, priv->counter_index, priv->port);
1281
1282         err = mlx4_en_config_rss_steer(priv);
1283         if (err) {
1284                 en_err(priv, "Failed configuring rss steering\n");
1285                 goto mac_err;
1286         }
1287
1288         err = mlx4_en_create_drop_qp(priv);
1289         if (err)
1290                 goto rss_err;
1291
1292         /* Configure tx cq's and rings */
1293         for (i = 0; i < priv->tx_ring_num; i++) {
1294                 /* Configure cq */
1295                 cq = priv->tx_cq[i];
1296                 err = mlx4_en_activate_cq(priv, cq, i);
1297                 if (err) {
1298                         en_err(priv, "Failed allocating Tx CQ\n");
1299                         goto tx_err;
1300                 }
1301                 err = mlx4_en_set_cq_moder(priv, cq);
1302                 if (err) {
1303                         en_err(priv, "Failed setting cq moderation parameters");
1304                         mlx4_en_deactivate_cq(priv, cq);
1305                         goto tx_err;
1306                 }
1307                 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1308                 cq->buf->wqe_index = cpu_to_be16(0xffff);
1309
1310                 /* Configure ring */
1311                 tx_ring = priv->tx_ring[i];
1312
1313                 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1314                                                i / priv->num_tx_rings_p_up);
1315                 if (err) {
1316                         en_err(priv, "Failed allocating Tx ring\n");
1317                         mlx4_en_deactivate_cq(priv, cq);
1318                         goto tx_err;
1319                 }
1320
1321                 /* Arm CQ for TX completions */
1322                 mlx4_en_arm_cq(priv, cq);
1323
1324                 /* Set initial ownership of all Tx TXBBs to SW (1) */
1325                 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1326                         *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1327                 ++tx_index;
1328         }
1329
1330         /* Configure port */
1331         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1332                                     priv->rx_mb_size,
1333                                     priv->prof->tx_pause,
1334                                     priv->prof->tx_ppp,
1335                                     priv->prof->rx_pause,
1336                                     priv->prof->rx_ppp);
1337         if (err) {
1338                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1339                        priv->port, err);
1340                 goto tx_err;
1341         }
1342         /* Set default qp number */
1343         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1344         if (err) {
1345                 en_err(priv, "Failed setting default qp numbers\n");
1346                 goto tx_err;
1347         }
1348
1349         /* Init port */
1350         en_dbg(HW, priv, "Initializing port\n");
1351         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1352         if (err) {
1353                 en_err(priv, "Failed Initializing port\n");
1354                 goto tx_err;
1355         }
1356
1357         /* Attach rx QP to bradcast address */
1358         memset(&mc_list[10], 0xff, ETH_ALEN);
1359         mc_list[5] = priv->port; /* needed for B0 steering support */
1360         if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1361                                   priv->port, 0, MLX4_PROT_ETH,
1362                                   &priv->broadcast_id))
1363                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1364
1365         /* Must redo promiscuous mode setup. */
1366         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1367
1368         /* Schedule multicast task to populate multicast list */
1369         queue_work(mdev->workqueue, &priv->rx_mode_task);
1370
1371         mlx4_set_stats_bitmap(mdev->dev, priv->stats_bitmap);
1372
1373         priv->port_up = true;
1374
1375         /* Enable the queues. */
1376         dev->if_drv_flags &= ~IFF_DRV_OACTIVE;
1377         dev->if_drv_flags |= IFF_DRV_RUNNING;
1378 #ifdef CONFIG_DEBUG_FS
1379         mlx4_en_create_debug_files(priv);
1380 #endif
1381         callout_reset(&priv->watchdog_timer, MLX4_EN_WATCHDOG_TIMEOUT,
1382                     mlx4_en_watchdog_timeout, priv);
1383
1384
1385         return 0;
1386
1387 tx_err:
1388         while (tx_index--) {
1389                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1390                 mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1391         }
1392         mlx4_en_destroy_drop_qp(priv);
1393 rss_err:
1394         mlx4_en_release_rss_steer(priv);
1395 mac_err:
1396         mlx4_en_put_qp(priv);
1397 cq_err:
1398         while (rx_index--)
1399                 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1400         for (i = 0; i < priv->rx_ring_num; i++)
1401                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1402
1403         return err; /* need to close devices */
1404 }
1405
1406
1407 void mlx4_en_stop_port(struct net_device *dev)
1408 {
1409         struct mlx4_en_priv *priv = netdev_priv(dev);
1410         struct mlx4_en_dev *mdev = priv->mdev;
1411         struct mlx4_en_mc_list *mclist, *tmp;
1412         int i;
1413         u8 mc_list[16] = {0};
1414
1415         if (!priv->port_up) {
1416                 en_dbg(DRV, priv, "stop port called while port already down\n");
1417                 return;
1418         }
1419
1420 #ifdef CONFIG_DEBUG_FS
1421         mlx4_en_delete_debug_files(priv);
1422 #endif
1423
1424         /* close port*/
1425         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1426
1427         /* Set port as not active */
1428         priv->port_up = false;
1429         if (priv->counter_index != 0xff) {
1430                 mlx4_counter_free(mdev->dev, priv->port, priv->counter_index);
1431                 priv->counter_index = 0xff;
1432         }
1433
1434         /* Promsicuous mode */
1435         if (mdev->dev->caps.steering_mode ==
1436             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1437                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1438                                  MLX4_EN_FLAG_MC_PROMISC);
1439                 mlx4_flow_steer_promisc_remove(mdev->dev,
1440                                                priv->port,
1441                                                MLX4_FS_ALL_DEFAULT);
1442                 mlx4_flow_steer_promisc_remove(mdev->dev,
1443                                                priv->port,
1444                                                MLX4_FS_MC_DEFAULT);
1445         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1446                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1447
1448                 /* Disable promiscouos mode */
1449                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1450                                             priv->port);
1451
1452                 /* Disable Multicast promisc */
1453                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1454                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1455                                                       priv->port);
1456                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1457                 }
1458         }
1459
1460         /* Detach All multicasts */
1461         memset(&mc_list[10], 0xff, ETH_ALEN);
1462         mc_list[5] = priv->port; /* needed for B0 steering support */
1463         mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1464                               MLX4_PROT_ETH, priv->broadcast_id);
1465         list_for_each_entry(mclist, &priv->curr_list, list) {
1466                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1467                 mc_list[5] = priv->port;
1468                 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1469                                       mc_list, MLX4_PROT_ETH, mclist->reg_id);
1470         }
1471         mlx4_en_clear_list(dev);
1472         list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1473                 list_del(&mclist->list);
1474                 kfree(mclist);
1475         }
1476
1477         /* Flush multicast filter */
1478         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1479         mlx4_en_destroy_drop_qp(priv);
1480
1481         /* Free TX Rings */
1482         for (i = 0; i < priv->tx_ring_num; i++) {
1483                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1484                 mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1485         }
1486         msleep(10);
1487
1488         for (i = 0; i < priv->tx_ring_num; i++)
1489                 mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1490
1491         /* Free RSS qps */
1492         mlx4_en_release_rss_steer(priv);
1493
1494         /* Unregister Mac address for the port */
1495         mlx4_en_put_qp(priv);
1496         mdev->mac_removed[priv->port] = 1;
1497
1498         /* Free RX Rings */
1499         for (i = 0; i < priv->rx_ring_num; i++) {
1500                 struct mlx4_en_cq *cq = priv->rx_cq[i];
1501                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1502                 mlx4_en_deactivate_cq(priv, cq);
1503         }
1504
1505         callout_stop(&priv->watchdog_timer);
1506
1507         dev->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1508 }
1509
1510 static void mlx4_en_restart(struct work_struct *work)
1511 {
1512         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1513                                                  watchdog_task);
1514         struct mlx4_en_dev *mdev = priv->mdev;
1515         struct net_device *dev = priv->dev;
1516         struct mlx4_en_tx_ring *ring;
1517         int i;
1518
1519
1520         if (priv->blocked == 0 || priv->port_up == 0)
1521                 return;
1522         for (i = 0; i < priv->tx_ring_num; i++) {
1523                 ring = priv->tx_ring[i];
1524                 if (ring->blocked &&
1525                                 ring->watchdog_time + MLX4_EN_WATCHDOG_TIMEOUT < ticks)
1526                         goto reset;
1527         }
1528         return;
1529
1530 reset:
1531         priv->port_stats.tx_timeout++;
1532         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1533
1534         mutex_lock(&mdev->state_lock);
1535         if (priv->port_up) {
1536                 mlx4_en_stop_port(dev);
1537                 //for (i = 0; i < priv->tx_ring_num; i++)
1538                 //        netdev_tx_reset_queue(priv->tx_ring[i]->tx_queue);
1539                 if (mlx4_en_start_port(dev))
1540                         en_err(priv, "Failed restarting port %d\n", priv->port);
1541         }
1542         mutex_unlock(&mdev->state_lock);
1543 }
1544
1545 static void mlx4_en_clear_stats(struct net_device *dev)
1546 {
1547         struct mlx4_en_priv *priv = netdev_priv(dev);
1548         struct mlx4_en_dev *mdev = priv->mdev;
1549         int i;
1550
1551         if (!mlx4_is_slave(mdev->dev))
1552                 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1553                         en_dbg(HW, priv, "Failed dumping statistics\n");
1554
1555         memset(&priv->pstats, 0, sizeof(priv->pstats));
1556         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1557         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1558         memset(&priv->vport_stats, 0, sizeof(priv->vport_stats));
1559
1560         for (i = 0; i < priv->tx_ring_num; i++) {
1561                 priv->tx_ring[i]->bytes = 0;
1562                 priv->tx_ring[i]->packets = 0;
1563                 priv->tx_ring[i]->tx_csum = 0;
1564         }
1565         for (i = 0; i < priv->rx_ring_num; i++) {
1566                 priv->rx_ring[i]->bytes = 0;
1567                 priv->rx_ring[i]->packets = 0;
1568                 priv->rx_ring[i]->csum_ok = 0;
1569                 priv->rx_ring[i]->csum_none = 0;
1570         }
1571 }
1572
1573 static void mlx4_en_open(void* arg)
1574 {
1575
1576         struct mlx4_en_priv *priv;
1577         struct mlx4_en_dev *mdev;
1578         struct net_device *dev;
1579         int err = 0;
1580
1581         priv = arg;
1582         mdev = priv->mdev;
1583         dev = priv->dev;
1584
1585
1586         mutex_lock(&mdev->state_lock);
1587
1588         if (!mdev->device_up) {
1589                 en_err(priv, "Cannot open - device down/disabled\n");
1590                 goto out;
1591         }
1592
1593         /* Reset HW statistics and SW counters */
1594         mlx4_en_clear_stats(dev);
1595
1596         err = mlx4_en_start_port(dev);
1597         if (err)
1598                 en_err(priv, "Failed starting port:%d\n", priv->port);
1599
1600 out:
1601         mutex_unlock(&mdev->state_lock);
1602         return;
1603 }
1604
1605 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1606 {
1607         int i;
1608
1609 #ifdef CONFIG_RFS_ACCEL
1610         if (priv->dev->rx_cpu_rmap) {
1611                 free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1612                 priv->dev->rx_cpu_rmap = NULL;
1613         }
1614 #endif
1615
1616         for (i = 0; i < priv->tx_ring_num; i++) {
1617                 if (priv->tx_ring && priv->tx_ring[i])
1618                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1619                 if (priv->tx_cq && priv->tx_cq[i])
1620                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1621         }
1622
1623         for (i = 0; i < priv->rx_ring_num; i++) {
1624                 if (priv->rx_ring[i])
1625                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1626                                 priv->prof->rx_ring_size, priv->stride);
1627                 if (priv->rx_cq[i])
1628                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1629         }
1630
1631         if (priv->sysctl)
1632                 sysctl_ctx_free(&priv->stat_ctx);
1633
1634
1635 }
1636
1637 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1638 {
1639         struct mlx4_en_port_profile *prof = priv->prof;
1640         int i;
1641         int node = 0;
1642
1643         /* Create rx Rings */
1644         for (i = 0; i < priv->rx_ring_num; i++) {
1645                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
1646                                       prof->rx_ring_size, i, RX, node))
1647                         goto err;
1648
1649                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
1650                                            prof->rx_ring_size, node))
1651                         goto err;
1652         }
1653
1654         /* Create tx Rings */
1655         for (i = 0; i < priv->tx_ring_num; i++) {
1656                 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1657                                       prof->tx_ring_size, i, TX, node))
1658                         goto err;
1659
1660                 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
1661                                            prof->tx_ring_size, TXBB_SIZE, node, i))
1662                         goto err;
1663         }
1664
1665 #ifdef CONFIG_RFS_ACCEL
1666         priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->rx_ring_num);
1667         if (!priv->dev->rx_cpu_rmap)
1668                 goto err;
1669 #endif
1670         /* Re-create stat sysctls in case the number of rings changed. */
1671         mlx4_en_sysctl_stat(priv);
1672         return 0;
1673
1674 err:
1675         en_err(priv, "Failed to allocate NIC resources\n");
1676         for (i = 0; i < priv->rx_ring_num; i++) {
1677                 if (priv->rx_ring[i])
1678                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1679                                                 prof->rx_ring_size,
1680                                                 priv->stride);
1681                 if (priv->rx_cq[i])
1682                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1683         }
1684         for (i = 0; i < priv->tx_ring_num; i++) {
1685                 if (priv->tx_ring[i])
1686                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1687                 if (priv->tx_cq[i])
1688                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1689         }
1690         priv->port_up = false;
1691         return -ENOMEM;
1692 }
1693
1694 struct en_port_attribute {
1695         struct attribute attr;
1696         ssize_t (*show)(struct en_port *, struct en_port_attribute *, char *buf);
1697         ssize_t (*store)(struct en_port *, struct en_port_attribute *, char *buf, size_t count);
1698 };
1699
1700 #define PORT_ATTR_RO(_name) \
1701 struct en_port_attribute en_port_attr_##_name = __ATTR_RO(_name)
1702
1703 #define EN_PORT_ATTR(_name, _mode, _show, _store) \
1704 struct en_port_attribute en_port_attr_##_name = __ATTR(_name, _mode, _show, _store)
1705
1706 void mlx4_en_destroy_netdev(struct net_device *dev)
1707 {
1708         struct mlx4_en_priv *priv = netdev_priv(dev);
1709         struct mlx4_en_dev *mdev = priv->mdev;
1710
1711         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
1712
1713         if (priv->vlan_attach != NULL)
1714                 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
1715         if (priv->vlan_detach != NULL)
1716                 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
1717
1718         /* Unregister device - this will close the port if it was up */
1719         if (priv->registered)
1720                 ether_ifdetach(dev);
1721
1722         if (priv->allocated)
1723                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
1724
1725         mutex_lock(&mdev->state_lock);
1726         mlx4_en_stop_port(dev);
1727         mutex_unlock(&mdev->state_lock);
1728
1729
1730         cancel_delayed_work(&priv->stats_task);
1731         cancel_delayed_work(&priv->service_task);
1732         /* flush any pending task for this netdev */
1733         flush_workqueue(mdev->workqueue);
1734         callout_drain(&priv->watchdog_timer);
1735
1736         /* Detach the netdev so tasks would not attempt to access it */
1737         mutex_lock(&mdev->state_lock);
1738         mdev->pndev[priv->port] = NULL;
1739         mutex_unlock(&mdev->state_lock);
1740
1741
1742         mlx4_en_free_resources(priv);
1743
1744         /* freeing the sysctl conf cannot be called from within mlx4_en_free_resources */
1745         if (priv->sysctl)
1746                 sysctl_ctx_free(&priv->conf_ctx);
1747
1748         kfree(priv->tx_ring);
1749         kfree(priv->tx_cq);
1750
1751         kfree(priv);
1752         if_free(dev);
1753
1754 }
1755
1756 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
1757 {
1758         struct mlx4_en_priv *priv = netdev_priv(dev);
1759         struct mlx4_en_dev *mdev = priv->mdev;
1760         int err = 0;
1761
1762         en_dbg(DRV, priv, "Change MTU called - current:%u new:%u\n",
1763                (unsigned)dev->if_mtu, (unsigned)new_mtu);
1764
1765         if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
1766                 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
1767                 return -EPERM;
1768         }
1769         mutex_lock(&mdev->state_lock);
1770         dev->if_mtu = new_mtu;
1771         if (dev->if_drv_flags & IFF_DRV_RUNNING) {
1772                 if (!mdev->device_up) {
1773                         /* NIC is probably restarting - let watchdog task reset
1774                          *                          * the port */
1775                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
1776                 } else {
1777                         mlx4_en_stop_port(dev);
1778                         err = mlx4_en_start_port(dev);
1779                         if (err) {
1780                                 en_err(priv, "Failed restarting port:%d\n",
1781                                                 priv->port);
1782                                 queue_work(mdev->workqueue, &priv->watchdog_task);
1783                         }
1784                 }
1785         }
1786         mutex_unlock(&mdev->state_lock);
1787         return 0;
1788 }
1789
1790 static int mlx4_en_calc_media(struct mlx4_en_priv *priv)
1791 {
1792         int trans_type;
1793         int active;
1794
1795         active = IFM_ETHER;
1796         if (priv->last_link_state == MLX4_DEV_EVENT_PORT_DOWN)
1797                 return (active);
1798         /*
1799          * [ShaharK] mlx4_en_QUERY_PORT sleeps and cannot be called under a
1800          * non-sleepable lock.
1801          * I moved it to the periodic mlx4_en_do_get_stats.
1802         if (mlx4_en_QUERY_PORT(priv->mdev, priv->port))
1803                 return (active);
1804         */
1805         active |= IFM_FDX;
1806         trans_type = priv->port_state.transciver;
1807         /* XXX I don't know all of the transceiver values. */
1808         switch (priv->port_state.link_speed) {
1809         case 1000:
1810                 active |= IFM_1000_T;
1811                 break;
1812         case 10000:
1813                 if (trans_type > 0 && trans_type <= 0xC)
1814                         active |= IFM_10G_SR;
1815                 else if (trans_type == 0x80 || trans_type == 0)
1816                         active |= IFM_10G_CX4;
1817                 break;
1818         case 40000:
1819                 active |= IFM_40G_CR4;
1820                 break;
1821         }
1822         if (priv->prof->tx_pause)
1823                 active |= IFM_ETH_TXPAUSE;
1824         if (priv->prof->rx_pause)
1825                 active |= IFM_ETH_RXPAUSE;
1826
1827         return (active);
1828 }
1829
1830 static void mlx4_en_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
1831 {
1832         struct mlx4_en_priv *priv;
1833
1834         priv = dev->if_softc;
1835         ifmr->ifm_status = IFM_AVALID;
1836         if (priv->last_link_state != MLX4_DEV_EVENT_PORT_DOWN)
1837                 ifmr->ifm_status |= IFM_ACTIVE;
1838         ifmr->ifm_active = mlx4_en_calc_media(priv);
1839
1840         return;
1841 }
1842
1843 static int mlx4_en_media_change(struct ifnet *dev)
1844 {
1845         struct mlx4_en_priv *priv;
1846         struct ifmedia *ifm;
1847         int rxpause;
1848         int txpause;
1849         int error;
1850
1851         priv = dev->if_softc;
1852         ifm = &priv->media;
1853         rxpause = txpause = 0;
1854         error = 0;
1855
1856         if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1857                 return (EINVAL);
1858         switch (IFM_SUBTYPE(ifm->ifm_media)) {
1859         case IFM_AUTO:
1860                 break;
1861         case IFM_10G_SR:
1862         case IFM_10G_CX4:
1863         case IFM_1000_T:
1864         case IFM_40G_CR4:
1865                 if ((IFM_SUBTYPE(ifm->ifm_media)
1866                         == IFM_SUBTYPE(mlx4_en_calc_media(priv)))
1867                         && (ifm->ifm_media & IFM_FDX))
1868                         break;
1869                 /* Fallthrough */
1870         default:
1871                 printf("%s: Only auto media type\n", if_name(dev));
1872                 return (EINVAL);
1873         }
1874         /* Allow user to set/clear pause */
1875         if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
1876                 rxpause = 1;
1877         if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
1878                 txpause = 1;
1879         if (priv->prof->tx_pause != txpause || priv->prof->rx_pause != rxpause) {
1880                 priv->prof->tx_pause = txpause;
1881                 priv->prof->rx_pause = rxpause;
1882                 error = -mlx4_SET_PORT_general(priv->mdev->dev, priv->port,
1883                      priv->rx_mb_size + ETHER_CRC_LEN, priv->prof->tx_pause,
1884                      priv->prof->tx_ppp, priv->prof->rx_pause,
1885                      priv->prof->rx_ppp);
1886         }
1887         return (error);
1888 }
1889
1890 static int mlx4_en_ioctl(struct ifnet *dev, u_long command, caddr_t data)
1891 {
1892         struct mlx4_en_priv *priv;
1893         struct mlx4_en_dev *mdev;
1894         struct ifreq *ifr;
1895         int error;
1896         int mask;
1897
1898         error = 0;
1899         mask = 0;
1900         priv = dev->if_softc;
1901         mdev = priv->mdev;
1902         ifr = (struct ifreq *) data;
1903         switch (command) {
1904
1905         case SIOCSIFMTU:
1906                 error = -mlx4_en_change_mtu(dev, ifr->ifr_mtu);
1907                 break;
1908         case SIOCSIFFLAGS:
1909                 mutex_lock(&mdev->state_lock);
1910                 if (dev->if_flags & IFF_UP) {
1911                         if ((dev->if_drv_flags & IFF_DRV_RUNNING) == 0)
1912                                 mlx4_en_start_port(dev);
1913                         else
1914                                 mlx4_en_set_rx_mode(dev);
1915                 } else {
1916                         if (dev->if_drv_flags & IFF_DRV_RUNNING) {
1917                                 mlx4_en_stop_port(dev);
1918                                 if_link_state_change(dev, LINK_STATE_DOWN);
1919                         }
1920                 }
1921                 mutex_unlock(&mdev->state_lock);
1922                 break;
1923         case SIOCADDMULTI:
1924         case SIOCDELMULTI:
1925                 mlx4_en_set_rx_mode(dev);
1926                 break;
1927         case SIOCSIFMEDIA:
1928         case SIOCGIFMEDIA:
1929                 error = ifmedia_ioctl(dev, ifr, &priv->media, command);
1930                 break;
1931         case SIOCSIFCAP:
1932                 mutex_lock(&mdev->state_lock);
1933                 mask = ifr->ifr_reqcap ^ dev->if_capenable;
1934                 if (mask & IFCAP_HWCSUM)
1935                         dev->if_capenable ^= IFCAP_HWCSUM;
1936                 if (mask & IFCAP_TSO4)
1937                         dev->if_capenable ^= IFCAP_TSO4;
1938                 if (mask & IFCAP_TSO6)
1939                         dev->if_capenable ^= IFCAP_TSO6;
1940                 if (mask & IFCAP_LRO)
1941                         dev->if_capenable ^= IFCAP_LRO;
1942                 if (mask & IFCAP_VLAN_HWTAGGING)
1943                         dev->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1944                 if (mask & IFCAP_VLAN_HWFILTER)
1945                         dev->if_capenable ^= IFCAP_VLAN_HWFILTER;
1946                 if (mask & IFCAP_WOL_MAGIC)
1947                         dev->if_capenable ^= IFCAP_WOL_MAGIC;
1948                 if (dev->if_drv_flags & IFF_DRV_RUNNING)
1949                         mlx4_en_start_port(dev);
1950                 mutex_unlock(&mdev->state_lock);
1951                 VLAN_CAPABILITIES(dev);
1952                 break;
1953         default:
1954                 error = ether_ioctl(dev, command, data);
1955                 break;
1956         }
1957
1958         return (error);
1959 }
1960
1961
1962 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
1963                         struct mlx4_en_port_profile *prof)
1964 {
1965         struct net_device *dev;
1966         struct mlx4_en_priv *priv;
1967         uint8_t dev_addr[ETHER_ADDR_LEN];
1968         int err;
1969         int i;
1970
1971         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1972         dev = priv->dev = if_alloc(IFT_ETHER);
1973         if (dev == NULL) {
1974                 en_err(priv, "Net device allocation failed\n");
1975                 kfree(priv);
1976                 return -ENOMEM;
1977         }
1978         dev->if_softc = priv;
1979         if_initname(dev, "mlxen", atomic_fetchadd_int(&mlx4_en_unit, 1));
1980         dev->if_mtu = ETHERMTU;
1981         dev->if_baudrate = 1000000000;
1982         dev->if_init = mlx4_en_open;
1983         dev->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1984         dev->if_ioctl = mlx4_en_ioctl;
1985         dev->if_transmit = mlx4_en_transmit;
1986         dev->if_qflush = mlx4_en_qflush;
1987         dev->if_snd.ifq_maxlen = prof->tx_ring_size;
1988
1989         /*
1990          * Initialize driver private data
1991          */
1992         priv->counter_index = 0xff;
1993         spin_lock_init(&priv->stats_lock);
1994         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
1995         INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
1996         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
1997         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
1998         INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
1999         callout_init(&priv->watchdog_timer, 1);
2000 #ifdef CONFIG_RFS_ACCEL
2001         INIT_LIST_HEAD(&priv->filters);
2002         spin_lock_init(&priv->filters_lock);
2003 #endif
2004
2005         priv->msg_enable = MLX4_EN_MSG_LEVEL;
2006         priv->dev = dev;
2007         priv->mdev = mdev;
2008         priv->ddev = &mdev->pdev->dev;
2009         priv->prof = prof;
2010         priv->port = port;
2011         priv->port_up = false;
2012         priv->flags = prof->flags;
2013         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2014                         MLX4_WQE_CTRL_SOLICITED);
2015
2016         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2017         priv->tx_ring_num = prof->tx_ring_num;
2018         priv->tx_ring = kcalloc(MAX_TX_RINGS,
2019                                 sizeof(struct mlx4_en_tx_ring *), GFP_KERNEL);
2020         if (!priv->tx_ring) {
2021                 err = -ENOMEM;
2022                 goto out;
2023         }
2024         priv->tx_cq = kcalloc(sizeof(struct mlx4_en_cq *), MAX_TX_RINGS,
2025                         GFP_KERNEL);
2026         if (!priv->tx_cq) {
2027                 err = -ENOMEM;
2028                 goto out;
2029         }
2030
2031         priv->rx_ring_num = prof->rx_ring_num;
2032         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2033         priv->mac_index = -1;
2034         priv->last_ifq_jiffies = 0;
2035         priv->if_counters_rx_errors = 0;
2036         priv->if_counters_rx_no_buffer = 0;
2037 #ifdef CONFIG_MLX4_EN_DCB
2038         if (!mlx4_is_slave(priv->mdev->dev)) {
2039                 priv->dcbx_cap = DCB_CAP_DCBX_HOST;
2040                 priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
2041                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
2042                         dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2043                 } else {
2044                         en_info(priv, "QoS disabled - no HW support\n");
2045                         dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2046                 }
2047         }
2048 #endif
2049
2050         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2051                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2052
2053
2054         /* Query for default mac and max mtu */
2055         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2056         priv->mac = mdev->dev->caps.def_mac[priv->port];
2057         if (ILLEGAL_MAC(priv->mac)) {
2058 #if BITS_PER_LONG == 64
2059                 en_err(priv, "Port: %d, invalid mac burned: 0x%lx, quiting\n",
2060                                 priv->port, priv->mac);
2061 #elif BITS_PER_LONG == 32
2062                 en_err(priv, "Port: %d, invalid mac burned: 0x%llx, quiting\n",
2063                                 priv->port, priv->mac);
2064 #endif
2065                 err = -EINVAL;
2066                 goto out;
2067         }
2068
2069
2070
2071         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2072                                           DS_SIZE);
2073
2074         mlx4_en_sysctl_conf(priv);
2075
2076         err = mlx4_en_alloc_resources(priv);
2077         if (err)
2078                 goto out;
2079
2080         /* Allocate page for receive rings */
2081         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2082                                 MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2083         if (err) {
2084                 en_err(priv, "Failed to allocate page for rx qps\n");
2085                 goto out;
2086         }
2087         priv->allocated = 1;
2088
2089         /*
2090          * Set driver features
2091          */
2092         dev->if_capabilities |= IFCAP_RXCSUM | IFCAP_TXCSUM;
2093         dev->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
2094         dev->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
2095         dev->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
2096         dev->if_capabilities |= IFCAP_LRO;
2097
2098         if (mdev->LSO_support)
2099                 dev->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_VLAN_HWTSO;
2100 #if 0
2101         /* set TSO limits so that we don't have to drop TX packets */
2102         dev->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2103         dev->if_hw_tsomaxsegcount = 16;
2104         dev->if_hw_tsomaxsegsize = 65536;       /* XXX can do up to 4GByte */
2105 #endif
2106         dev->if_capenable = dev->if_capabilities;
2107
2108         dev->if_hwassist = 0;
2109         if (dev->if_capenable & (IFCAP_TSO4 | IFCAP_TSO6))
2110                 dev->if_hwassist |= CSUM_TSO;
2111         if (dev->if_capenable & IFCAP_TXCSUM)
2112                 dev->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2113
2114
2115         /* Register for VLAN events */
2116         priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
2117             mlx4_en_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
2118         priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
2119             mlx4_en_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
2120
2121         mdev->pndev[priv->port] = dev;
2122
2123         priv->last_link_state = MLX4_DEV_EVENT_PORT_DOWN;
2124         mlx4_en_set_default_moderation(priv);
2125
2126         /* Set default MAC */
2127         for (i = 0; i < ETHER_ADDR_LEN; i++)
2128                 dev_addr[ETHER_ADDR_LEN - 1 - i] = (u8) (priv->mac >> (8 * i));
2129
2130
2131         ether_ifattach(dev, dev_addr);
2132         if_link_state_change(dev, LINK_STATE_DOWN);
2133         ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
2134             mlx4_en_media_change, mlx4_en_media_status);
2135         ifmedia_add(&priv->media, IFM_ETHER | IFM_FDX | IFM_1000_T, 0, NULL);
2136         ifmedia_add(&priv->media, IFM_ETHER | IFM_FDX | IFM_10G_SR, 0, NULL);
2137         ifmedia_add(&priv->media, IFM_ETHER | IFM_FDX | IFM_10G_CX4, 0, NULL);
2138         ifmedia_add(&priv->media, IFM_ETHER | IFM_FDX | IFM_40G_CR4, 0, NULL);
2139         ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2140         ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO);
2141
2142         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2143         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2144
2145         priv->registered = 1;
2146
2147         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2148         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2149
2150
2151         priv->rx_mb_size = dev->if_mtu + ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN;
2152         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2153                                     priv->rx_mb_size,
2154                                     prof->tx_pause, prof->tx_ppp,
2155                                     prof->rx_pause, prof->rx_ppp);
2156         if (err) {
2157                 en_err(priv, "Failed setting port general configurations "
2158                        "for port %d, with error %d\n", priv->port, err);
2159                 goto out;
2160         }
2161
2162         /* Init port */
2163         en_warn(priv, "Initializing port\n");
2164         err = mlx4_INIT_PORT(mdev->dev, priv->port);
2165         if (err) {
2166                 en_err(priv, "Failed Initializing port\n");
2167                 goto out;
2168         }
2169
2170         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2171
2172         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2173                 queue_delayed_work(mdev->workqueue, &priv->service_task, SERVICE_TASK_DELAY);
2174
2175
2176
2177         return 0;
2178
2179 out:
2180         mlx4_en_destroy_netdev(dev);
2181         return err;
2182 }
2183 static int mlx4_en_set_ring_size(struct net_device *dev,
2184     int rx_size, int tx_size)
2185 {
2186         struct mlx4_en_priv *priv = netdev_priv(dev);
2187         struct mlx4_en_dev *mdev = priv->mdev;
2188         int port_up = 0;
2189         int err = 0;
2190
2191         rx_size = roundup_pow_of_two(rx_size);
2192         rx_size = max_t(u32, rx_size, MLX4_EN_MIN_RX_SIZE);
2193         rx_size = min_t(u32, rx_size, MLX4_EN_MAX_RX_SIZE);
2194         tx_size = roundup_pow_of_two(tx_size);
2195         tx_size = max_t(u32, tx_size, MLX4_EN_MIN_TX_SIZE);
2196         tx_size = min_t(u32, tx_size, MLX4_EN_MAX_TX_SIZE);
2197
2198         if (rx_size == (priv->port_up ?
2199             priv->rx_ring[0]->actual_size : priv->rx_ring[0]->size) &&
2200             tx_size == priv->tx_ring[0]->size)
2201                 return 0;
2202         mutex_lock(&mdev->state_lock);
2203         if (priv->port_up) {
2204                 port_up = 1;
2205                 mlx4_en_stop_port(dev);
2206         }
2207         mlx4_en_free_resources(priv);
2208         priv->prof->tx_ring_size = tx_size;
2209         priv->prof->rx_ring_size = rx_size;
2210         err = mlx4_en_alloc_resources(priv);
2211         if (err) {
2212                 en_err(priv, "Failed reallocating port resources\n");
2213                 goto out;
2214         }
2215         if (port_up) {
2216                 err = mlx4_en_start_port(dev);
2217                 if (err)
2218                         en_err(priv, "Failed starting port\n");
2219         }
2220 out:
2221         mutex_unlock(&mdev->state_lock);
2222         return err;
2223 }
2224 static int mlx4_en_set_rx_ring_size(SYSCTL_HANDLER_ARGS)
2225 {
2226         struct mlx4_en_priv *priv;
2227         int size;
2228         int error;
2229
2230         priv = arg1;
2231         size = priv->prof->rx_ring_size;
2232         error = sysctl_handle_int(oidp, &size, 0, req);
2233         if (error || !req->newptr)
2234                 return (error);
2235         error = -mlx4_en_set_ring_size(priv->dev, size,
2236             priv->prof->tx_ring_size);
2237         return (error);
2238 }
2239
2240 static int mlx4_en_set_tx_ring_size(SYSCTL_HANDLER_ARGS)
2241 {
2242         struct mlx4_en_priv *priv;
2243         int size;
2244         int error;
2245
2246         priv = arg1;
2247         size = priv->prof->tx_ring_size;
2248         error = sysctl_handle_int(oidp, &size, 0, req);
2249         if (error || !req->newptr)
2250                 return (error);
2251         error = -mlx4_en_set_ring_size(priv->dev, priv->prof->rx_ring_size,
2252             size);
2253
2254         return (error);
2255 }
2256
2257 static int mlx4_en_set_tx_ppp(SYSCTL_HANDLER_ARGS)
2258 {
2259         struct mlx4_en_priv *priv;
2260         int ppp;
2261         int error;
2262
2263         priv = arg1;
2264         ppp = priv->prof->tx_ppp;
2265         error = sysctl_handle_int(oidp, &ppp, 0, req);
2266         if (error || !req->newptr)
2267                 return (error);
2268         if (ppp > 0xff || ppp < 0)
2269                 return (-EINVAL);
2270         priv->prof->tx_ppp = ppp;
2271         error = -mlx4_SET_PORT_general(priv->mdev->dev, priv->port,
2272                                        priv->rx_mb_size + ETHER_CRC_LEN,
2273                                        priv->prof->tx_pause,
2274                                        priv->prof->tx_ppp,
2275                                        priv->prof->rx_pause,
2276                                        priv->prof->rx_ppp);
2277
2278         return (error);
2279 }
2280
2281 static int mlx4_en_set_rx_ppp(SYSCTL_HANDLER_ARGS)
2282 {
2283         struct mlx4_en_priv *priv;
2284         struct mlx4_en_dev *mdev;
2285         int ppp;
2286         int error;
2287         int port_up;
2288
2289         port_up = 0;
2290         priv = arg1;
2291         mdev = priv->mdev;
2292         ppp = priv->prof->rx_ppp;
2293         error = sysctl_handle_int(oidp, &ppp, 0, req);
2294         if (error || !req->newptr)
2295                 return (error);
2296         if (ppp > 0xff || ppp < 0)
2297                 return (-EINVAL);
2298         /* See if we have to change the number of tx queues. */
2299         if (!ppp != !priv->prof->rx_ppp) {
2300                 mutex_lock(&mdev->state_lock);
2301                 if (priv->port_up) {
2302                         port_up = 1;
2303                         mlx4_en_stop_port(priv->dev);
2304                 }
2305                 mlx4_en_free_resources(priv);
2306                 priv->prof->rx_ppp = ppp;
2307                 error = -mlx4_en_alloc_resources(priv);
2308                 if (error)
2309                         en_err(priv, "Failed reallocating port resources\n");
2310                 if (error == 0 && port_up) {
2311                         error = -mlx4_en_start_port(priv->dev);
2312                         if (error)
2313                                 en_err(priv, "Failed starting port\n");
2314                 }
2315                 mutex_unlock(&mdev->state_lock);
2316                 return (error);
2317
2318         }
2319         priv->prof->rx_ppp = ppp;
2320         error = -mlx4_SET_PORT_general(priv->mdev->dev, priv->port,
2321                                        priv->rx_mb_size + ETHER_CRC_LEN,
2322                                        priv->prof->tx_pause,
2323                                        priv->prof->tx_ppp,
2324                                        priv->prof->rx_pause,
2325                                        priv->prof->rx_ppp);
2326
2327         return (error);
2328 }
2329
2330 static void mlx4_en_sysctl_conf(struct mlx4_en_priv *priv)
2331 {
2332         struct net_device *dev;
2333         struct sysctl_ctx_list *ctx;
2334         struct sysctl_oid *node;
2335         struct sysctl_oid_list *node_list;
2336         struct sysctl_oid *coal;
2337         struct sysctl_oid_list *coal_list;
2338         const char *pnameunit;
2339
2340         dev = priv->dev;
2341         ctx = &priv->conf_ctx;
2342         pnameunit = device_get_nameunit(priv->mdev->pdev->dev.bsddev);
2343
2344         sysctl_ctx_init(ctx);
2345         priv->sysctl = SYSCTL_ADD_NODE(ctx, SYSCTL_STATIC_CHILDREN(_hw),
2346             OID_AUTO, dev->if_xname, CTLFLAG_RD, 0, "mlx4 10gig ethernet");
2347         node = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(priv->sysctl), OID_AUTO,
2348             "conf", CTLFLAG_RD, NULL, "Configuration");
2349         node_list = SYSCTL_CHILDREN(node);
2350
2351         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "msg_enable",
2352             CTLFLAG_RW, &priv->msg_enable, 0,
2353             "Driver message enable bitfield");
2354         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "rx_rings",
2355             CTLFLAG_RD, &priv->rx_ring_num, 0,
2356             "Number of receive rings");
2357         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "tx_rings",
2358             CTLFLAG_RD, &priv->tx_ring_num, 0,
2359             "Number of transmit rings");
2360         SYSCTL_ADD_PROC(ctx, node_list, OID_AUTO, "rx_size",
2361             CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, priv, 0,
2362             mlx4_en_set_rx_ring_size, "I", "Receive ring size");
2363         SYSCTL_ADD_PROC(ctx, node_list, OID_AUTO, "tx_size",
2364             CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, priv, 0,
2365             mlx4_en_set_tx_ring_size, "I", "Transmit ring size");
2366         SYSCTL_ADD_PROC(ctx, node_list, OID_AUTO, "tx_ppp",
2367             CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, priv, 0,
2368             mlx4_en_set_tx_ppp, "I", "TX Per-priority pause");
2369         SYSCTL_ADD_PROC(ctx, node_list, OID_AUTO, "rx_ppp",
2370             CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, priv, 0,
2371             mlx4_en_set_rx_ppp, "I", "RX Per-priority pause");
2372         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "port_num",
2373             CTLFLAG_RD, &priv->port, 0,
2374             "Port Number");
2375         SYSCTL_ADD_STRING(ctx, node_list, OID_AUTO, "device_name",
2376             CTLFLAG_RD, __DECONST(void *, pnameunit), 0,
2377             "PCI device name");
2378
2379         /* Add coalescer configuration. */
2380         coal = SYSCTL_ADD_NODE(ctx, node_list, OID_AUTO,
2381             "coalesce", CTLFLAG_RD, NULL, "Interrupt coalesce configuration");
2382         coal_list = SYSCTL_CHILDREN(node);
2383         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "pkt_rate_low",
2384             CTLFLAG_RW, &priv->pkt_rate_low, 0,
2385             "Packets per-second for minimum delay");
2386         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "rx_usecs_low",
2387             CTLFLAG_RW, &priv->rx_usecs_low, 0,
2388             "Minimum RX delay in micro-seconds");
2389         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "pkt_rate_high",
2390             CTLFLAG_RW, &priv->pkt_rate_high, 0,
2391             "Packets per-second for maximum delay");
2392         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "rx_usecs_high",
2393             CTLFLAG_RW, &priv->rx_usecs_high, 0,
2394             "Maximum RX delay in micro-seconds");
2395         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "sample_interval",
2396             CTLFLAG_RW, &priv->sample_interval, 0,
2397             "adaptive frequency in units of HZ ticks");
2398         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "adaptive_rx_coal",
2399             CTLFLAG_RW, &priv->adaptive_rx_coal, 0,
2400             "Enable adaptive rx coalescing");
2401 }
2402
2403
2404 static void mlx4_en_sysctl_stat(struct mlx4_en_priv *priv)
2405 {
2406         struct net_device *dev;
2407         struct sysctl_ctx_list *ctx;
2408         struct sysctl_oid *node;
2409         struct sysctl_oid_list *node_list;
2410         struct sysctl_oid *ring_node;
2411         struct sysctl_oid_list *ring_list;
2412         struct mlx4_en_tx_ring *tx_ring;
2413         struct mlx4_en_rx_ring *rx_ring;
2414         char namebuf[128];
2415         int i;
2416
2417         dev = priv->dev;
2418
2419         ctx = &priv->stat_ctx;
2420         sysctl_ctx_init(ctx);
2421         node = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(priv->sysctl), OID_AUTO,
2422             "stat", CTLFLAG_RD, NULL, "Statistics");
2423         node_list = SYSCTL_CHILDREN(node);
2424
2425 #ifdef MLX4_EN_PERF_STAT
2426         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "tx_poll", CTLFLAG_RD,
2427             &priv->pstats.tx_poll, "TX Poll calls");
2428         SYSCTL_ADD_QUAD(ctx, node_list, OID_AUTO, "tx_pktsz_avg", CTLFLAG_RD,
2429             &priv->pstats.tx_pktsz_avg, "TX average packet size");
2430         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "inflight_avg", CTLFLAG_RD,
2431             &priv->pstats.inflight_avg, "TX average packets in-flight");
2432         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "tx_coal_avg", CTLFLAG_RD,
2433             &priv->pstats.tx_coal_avg, "TX average coalesced completions");
2434         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "rx_coal_avg", CTLFLAG_RD,
2435             &priv->pstats.rx_coal_avg, "RX average coalesced completions");
2436 #endif
2437
2438         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tso_packets", CTLFLAG_RD,
2439             &priv->port_stats.tso_packets, "TSO packets sent");
2440         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "queue_stopped", CTLFLAG_RD,
2441             &priv->port_stats.queue_stopped, "Queue full");
2442         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "wake_queue", CTLFLAG_RD,
2443             &priv->port_stats.wake_queue, "Queue resumed after full");
2444         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_timeout", CTLFLAG_RD,
2445             &priv->port_stats.tx_timeout, "Transmit timeouts");
2446         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_alloc_failed", CTLFLAG_RD,
2447             &priv->port_stats.rx_alloc_failed, "RX failed to allocate mbuf");
2448         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_chksum_good", CTLFLAG_RD,
2449             &priv->port_stats.rx_chksum_good, "RX checksum offload success");
2450         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_chksum_none", CTLFLAG_RD,
2451             &priv->port_stats.rx_chksum_none, "RX without checksum offload");
2452         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_chksum_offload",
2453             CTLFLAG_RD, &priv->port_stats.tx_chksum_offload,
2454             "TX checksum offloads");
2455
2456         /* Could strdup the names and add in a loop.  This is simpler. */
2457         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_bytes", CTLFLAG_RD,
2458             &priv->pkstats.rx_bytes, "RX Bytes");
2459         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_packets", CTLFLAG_RD,
2460             &priv->pkstats.rx_packets, "RX packets");
2461         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_multicast_packets", CTLFLAG_RD,
2462             &priv->pkstats.rx_multicast_packets, "RX Multicast Packets");
2463         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_broadcast_packets", CTLFLAG_RD,
2464             &priv->pkstats.rx_broadcast_packets, "RX Broadcast Packets");
2465         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_errors", CTLFLAG_RD,
2466             &priv->pkstats.rx_errors, "RX Errors");
2467         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_dropped", CTLFLAG_RD,
2468             &priv->pkstats.rx_dropped, "RX Dropped");
2469         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_length_errors", CTLFLAG_RD,
2470             &priv->pkstats.rx_length_errors, "RX Length Errors");
2471         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_over_errors", CTLFLAG_RD,
2472             &priv->pkstats.rx_over_errors, "RX Over Errors");
2473         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_crc_errors", CTLFLAG_RD,
2474             &priv->pkstats.rx_crc_errors, "RX CRC Errors");
2475         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_jabbers", CTLFLAG_RD,
2476             &priv->pkstats.rx_jabbers, "RX Jabbers");
2477
2478
2479         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_in_range_length_error", CTLFLAG_RD,
2480             &priv->pkstats.rx_in_range_length_error, "RX IN_Range Length Error");
2481         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_out_range_length_error",
2482                 CTLFLAG_RD, &priv->pkstats.rx_out_range_length_error,
2483                 "RX Out Range Length Error");
2484         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_lt_64_bytes_packets", CTLFLAG_RD,
2485             &priv->pkstats.rx_lt_64_bytes_packets, "RX Lt 64 Bytes Packets");
2486         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_127_bytes_packets", CTLFLAG_RD,
2487             &priv->pkstats.rx_127_bytes_packets, "RX 127 bytes Packets");
2488         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_255_bytes_packets", CTLFLAG_RD,
2489             &priv->pkstats.rx_255_bytes_packets, "RX 255 bytes Packets");
2490         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_511_bytes_packets", CTLFLAG_RD,
2491             &priv->pkstats.rx_511_bytes_packets, "RX 511 bytes Packets");
2492         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_1023_bytes_packets", CTLFLAG_RD,
2493             &priv->pkstats.rx_1023_bytes_packets, "RX 1023 bytes Packets");
2494         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_1518_bytes_packets", CTLFLAG_RD,
2495             &priv->pkstats.rx_1518_bytes_packets, "RX 1518 bytes Packets");
2496         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_1522_bytes_packets", CTLFLAG_RD,
2497             &priv->pkstats.rx_1522_bytes_packets, "RX 1522 bytes Packets");
2498         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_1548_bytes_packets", CTLFLAG_RD,
2499             &priv->pkstats.rx_1548_bytes_packets, "RX 1548 bytes Packets");
2500         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_gt_1548_bytes_packets", CTLFLAG_RD,
2501             &priv->pkstats.rx_gt_1548_bytes_packets,
2502             "RX Greater Then 1548 bytes Packets");
2503
2504 struct mlx4_en_pkt_stats {
2505         unsigned long tx_packets;
2506         unsigned long tx_bytes;
2507         unsigned long tx_multicast_packets;
2508         unsigned long tx_broadcast_packets;
2509         unsigned long tx_errors;
2510         unsigned long tx_dropped;
2511         unsigned long tx_lt_64_bytes_packets;
2512         unsigned long tx_127_bytes_packets;
2513         unsigned long tx_255_bytes_packets;
2514         unsigned long tx_511_bytes_packets;
2515         unsigned long tx_1023_bytes_packets;
2516         unsigned long tx_1518_bytes_packets;
2517         unsigned long tx_1522_bytes_packets;
2518         unsigned long tx_1548_bytes_packets;
2519         unsigned long tx_gt_1548_bytes_packets;
2520         unsigned long rx_prio[NUM_PRIORITIES][NUM_PRIORITY_STATS];
2521         unsigned long tx_prio[NUM_PRIORITIES][NUM_PRIORITY_STATS];
2522 #define NUM_PKT_STATS           72
2523 };
2524
2525
2526         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_packets", CTLFLAG_RD,
2527             &priv->pkstats.tx_packets, "TX packets");
2528         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_bytes", CTLFLAG_RD,
2529             &priv->pkstats.tx_packets, "TX Bytes");
2530         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_multicast_packets", CTLFLAG_RD,
2531             &priv->pkstats.tx_multicast_packets, "TX Multicast Packets");
2532         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_broadcast_packets", CTLFLAG_RD,
2533             &priv->pkstats.tx_broadcast_packets, "TX Broadcast Packets");
2534         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_errors", CTLFLAG_RD,
2535             &priv->pkstats.tx_errors, "TX Errors");
2536         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_dropped", CTLFLAG_RD,
2537             &priv->pkstats.tx_dropped, "TX Dropped");
2538         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_lt_64_bytes_packets", CTLFLAG_RD,
2539             &priv->pkstats.tx_lt_64_bytes_packets, "TX Less Then 64 Bytes Packets");
2540         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_127_bytes_packets", CTLFLAG_RD,
2541             &priv->pkstats.tx_127_bytes_packets, "TX 127 Bytes Packets");
2542         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_255_bytes_packets", CTLFLAG_RD,
2543             &priv->pkstats.tx_255_bytes_packets, "TX 255 Bytes Packets");
2544         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_511_bytes_packets", CTLFLAG_RD,
2545             &priv->pkstats.tx_511_bytes_packets, "TX 511 Bytes Packets");
2546         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_1023_bytes_packets", CTLFLAG_RD,
2547             &priv->pkstats.tx_1023_bytes_packets, "TX 1023 Bytes Packets");
2548         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_1518_bytes_packets", CTLFLAG_RD,
2549             &priv->pkstats.tx_1518_bytes_packets, "TX 1518 Bytes Packets");
2550         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_1522_bytes_packets", CTLFLAG_RD,
2551             &priv->pkstats.tx_1522_bytes_packets, "TX 1522 Bytes Packets");
2552         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_1548_bytes_packets", CTLFLAG_RD,
2553             &priv->pkstats.tx_1548_bytes_packets, "TX 1548 Bytes Packets");
2554         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_gt_1548_bytes_packets", CTLFLAG_RD,
2555             &priv->pkstats.tx_gt_1548_bytes_packets,
2556             "TX Greater Then 1548 Bytes Packets");
2557
2558
2559
2560         for (i = 0; i < priv->tx_ring_num; i++) {
2561                 tx_ring = priv->tx_ring[i];
2562                 snprintf(namebuf, sizeof(namebuf), "tx_ring%d", i);
2563                 ring_node = SYSCTL_ADD_NODE(ctx, node_list, OID_AUTO, namebuf,
2564                     CTLFLAG_RD, NULL, "TX Ring");
2565                 ring_list = SYSCTL_CHILDREN(ring_node);
2566                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "packets",
2567                     CTLFLAG_RD, &tx_ring->packets, "TX packets");
2568                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "bytes",
2569                     CTLFLAG_RD, &tx_ring->bytes, "TX bytes");
2570
2571         }
2572         for (i = 0; i < priv->rx_ring_num; i++) {
2573                 rx_ring = priv->rx_ring[i];
2574                 snprintf(namebuf, sizeof(namebuf), "rx_ring%d", i);
2575                 ring_node = SYSCTL_ADD_NODE(ctx, node_list, OID_AUTO, namebuf,
2576                     CTLFLAG_RD, NULL, "RX Ring");
2577                 ring_list = SYSCTL_CHILDREN(ring_node);
2578                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "packets",
2579                     CTLFLAG_RD, &rx_ring->packets, "RX packets");
2580                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "bytes",
2581                     CTLFLAG_RD, &rx_ring->bytes, "RX bytes");
2582                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "error",
2583                     CTLFLAG_RD, &rx_ring->errors, "RX soft errors");
2584         }
2585 }