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         if_maddr_rlock(dev);
 654         TAILQ_FOREACH(ifma, &dev->if_multiaddrs, ifma_link) {
 655                 if (ifma->ifma_addr->sa_family != AF_LINK)
 656                         continue;
 657                 if (((struct sockaddr_dl *)ifma->ifma_addr)->sdl_alen !=
 658                                 ETHER_ADDR_LEN)
 659                         continue;
 660                 /* Make sure the list didn't grow. */
 661                 tmp = kzalloc(sizeof(struct mlx4_en_mc_list), GFP_ATOMIC);
 662                 if (tmp == NULL)
 663                         break;
 664                 memcpy(tmp->addr,
 665                         LLADDR((struct sockaddr_dl *)ifma->ifma_addr), ETH_ALEN);
 666                 list_add_tail(&tmp->list, &priv->mc_list);
 667         }
 668         if_maddr_runlock(dev);
 669 }
 670
 671 static void update_mclist_flags(struct mlx4_en_priv *priv,
 672                                 struct list_head *dst,
 673                                 struct list_head *src)
 674 {
 675         struct mlx4_en_mc_list *dst_tmp, *src_tmp, *new_mc;
 676         bool found;
 677
 678         /* Find all the entries that should be removed from dst,
 679          * These are the entries that are not found in src
 680          */
 681         list_for_each_entry(dst_tmp, dst, list) {
 682                 found = false;
 683                 list_for_each_entry(src_tmp, src, list) {
 684                         if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
 685                                 found = true;
 686                                 break;
 687                         }
 688                 }
 689                 if (!found)
 690                         dst_tmp->action = MCLIST_REM;
 691         }
 692
 693         /* Add entries that exist in src but not in dst
 694          * mark them as need to add
 695          */
 696         list_for_each_entry(src_tmp, src, list) {
 697                 found = false;
 698                 list_for_each_entry(dst_tmp, dst, list) {
 699                         if (!memcmp(dst_tmp->addr, src_tmp->addr, ETH_ALEN)) {
 700                                 dst_tmp->action = MCLIST_NONE;
 701                                 found = true;
 702                                 break;
 703                         }
 704                 }
 705                 if (!found) {
 706                         new_mc = kmalloc(sizeof(struct mlx4_en_mc_list),
 707                                          GFP_KERNEL);
 708                         if (!new_mc) {
 709                                 en_err(priv, "Failed to allocate current multicast list\n");
 710                                 return;
 711                         }
 712                         memcpy(new_mc, src_tmp,
 713                                sizeof(struct mlx4_en_mc_list));
 714                         new_mc->action = MCLIST_ADD;
 715                         list_add_tail(&new_mc->list, dst);
 716                 }
 717         }
 718 }
 719
 720 static void mlx4_en_set_rx_mode(struct net_device *dev)
 721 {
 722         struct mlx4_en_priv *priv = netdev_priv(dev);
 723
 724         if (!priv->port_up)
 725                 return;
 726
 727         queue_work(priv->mdev->workqueue, &priv->rx_mode_task);
 728 }
 729
 730 static void mlx4_en_set_promisc_mode(struct mlx4_en_priv *priv,
 731                                      struct mlx4_en_dev *mdev)
 732 {
 733         int err = 0;
 734         if (!(priv->flags & MLX4_EN_FLAG_PROMISC)) {
 735                 priv->flags |= MLX4_EN_FLAG_PROMISC;
 736
 737                 /* Enable promiscouos mode */
 738                 switch (mdev->dev->caps.steering_mode) {
 739                 case MLX4_STEERING_MODE_DEVICE_MANAGED:
 740                         err = mlx4_flow_steer_promisc_add(mdev->dev,
 741                                                           priv->port,
 742                                                           priv->base_qpn,
 743                                                           MLX4_FS_ALL_DEFAULT);
 744                         if (err)
 745                                 en_err(priv, "Failed enabling promiscuous mode\n");
 746                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 747                         break;
 748
 749                 case MLX4_STEERING_MODE_B0:
 750                         err = mlx4_unicast_promisc_add(mdev->dev,
 751                                                        priv->base_qpn,
 752                                                        priv->port);
 753                         if (err)
 754                                 en_err(priv, "Failed enabling unicast promiscuous mode\n");
 755
 756                         /* Add the default qp number as multicast
 757                          * promisc
 758                          */
 759                         if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 760                                 err = mlx4_multicast_promisc_add(mdev->dev,
 761                                                                  priv->base_qpn,
 762                                                                  priv->port);
 763                                 if (err)
 764                                         en_err(priv, "Failed enabling multicast promiscuous mode\n");
 765                                 priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 766                         }
 767                         break;
 768
 769                 case MLX4_STEERING_MODE_A0:
 770                         err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 771                                                      priv->port,
 772                                                      priv->base_qpn,
 773                                                      1);
 774                         if (err)
 775                                 en_err(priv, "Failed enabling promiscuous mode\n");
 776                         break;
 777                 }
 778
 779                 /* Disable port multicast filter (unconditionally) */
 780                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 781                                           0, MLX4_MCAST_DISABLE);
 782                 if (err)
 783                         en_err(priv, "Failed disabling multicast filter\n");
 784         }
 785 }
 786
 787 static void mlx4_en_clear_promisc_mode(struct mlx4_en_priv *priv,
 788                                        struct mlx4_en_dev *mdev)
 789 {
 790         int err = 0;
 791
 792         priv->flags &= ~MLX4_EN_FLAG_PROMISC;
 793
 794         /* Disable promiscouos mode */
 795         switch (mdev->dev->caps.steering_mode) {
 796         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 797                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 798                                                      priv->port,
 799                                                      MLX4_FS_ALL_DEFAULT);
 800                 if (err)
 801                         en_err(priv, "Failed disabling promiscuous mode\n");
 802                 priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 803                 break;
 804
 805         case MLX4_STEERING_MODE_B0:
 806                 err = mlx4_unicast_promisc_remove(mdev->dev,
 807                                                   priv->base_qpn,
 808                                                   priv->port);
 809                 if (err)
 810                         en_err(priv, "Failed disabling unicast promiscuous mode\n");
 811                 /* Disable Multicast promisc */
 812                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 813                         err = mlx4_multicast_promisc_remove(mdev->dev,
 814                                                             priv->base_qpn,
 815                                                             priv->port);
 816                         if (err)
 817                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 818                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 819                 }
 820                 break;
 821
 822         case MLX4_STEERING_MODE_A0:
 823                 err = mlx4_SET_PORT_qpn_calc(mdev->dev,
 824                                              priv->port,
 825                                              priv->base_qpn, 0);
 826                 if (err)
 827                         en_err(priv, "Failed disabling promiscuous mode\n");
 828                 break;
 829         }
 830 }
 831
 832 static void mlx4_en_do_multicast(struct mlx4_en_priv *priv,
 833                                  struct net_device *dev,
 834                                  struct mlx4_en_dev *mdev)
 835 {
 836         struct mlx4_en_mc_list *mclist, *tmp;
 837         u8 mc_list[16] = {0};
 838         int err = 0;
 839         u64 mcast_addr = 0;
 840
 841
 842         /* Enable/disable the multicast filter according to IFF_ALLMULTI */
 843         if (dev->if_flags & IFF_ALLMULTI) {
 844                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 845                                           0, MLX4_MCAST_DISABLE);
 846                 if (err)
 847                         en_err(priv, "Failed disabling multicast filter\n");
 848
 849                 /* Add the default qp number as multicast promisc */
 850                 if (!(priv->flags & MLX4_EN_FLAG_MC_PROMISC)) {
 851                         switch (mdev->dev->caps.steering_mode) {
 852                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 853                                 err = mlx4_flow_steer_promisc_add(mdev->dev,
 854                                                                   priv->port,
 855                                                                   priv->base_qpn,
 856                                                                   MLX4_FS_MC_DEFAULT);
 857                                 break;
 858
 859                         case MLX4_STEERING_MODE_B0:
 860                                 err = mlx4_multicast_promisc_add(mdev->dev,
 861                                                                  priv->base_qpn,
 862                                                                  priv->port);
 863                                 break;
 864
 865                         case MLX4_STEERING_MODE_A0:
 866                                 break;
 867                         }
 868                         if (err)
 869                                 en_err(priv, "Failed entering multicast promisc mode\n");
 870                         priv->flags |= MLX4_EN_FLAG_MC_PROMISC;
 871                 }
 872         } else {
 873                 /* Disable Multicast promisc */
 874                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
 875                         switch (mdev->dev->caps.steering_mode) {
 876                         case MLX4_STEERING_MODE_DEVICE_MANAGED:
 877                                 err = mlx4_flow_steer_promisc_remove(mdev->dev,
 878                                                                      priv->port,
 879                                                                      MLX4_FS_MC_DEFAULT);
 880                                 break;
 881
 882                         case MLX4_STEERING_MODE_B0:
 883                                 err = mlx4_multicast_promisc_remove(mdev->dev,
 884                                                                     priv->base_qpn,
 885                                                                     priv->port);
 886                                 break;
 887
 888                         case MLX4_STEERING_MODE_A0:
 889                                 break;
 890                         }
 891                         if (err)
 892                                 en_err(priv, "Failed disabling multicast promiscuous mode\n");
 893                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
 894                 }
 895
 896                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 897                                           0, MLX4_MCAST_DISABLE);
 898                 if (err)
 899                         en_err(priv, "Failed disabling multicast filter\n");
 900
 901                 /* Flush mcast filter and init it with broadcast address */
 902                 mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, ETH_BCAST,
 903                                     1, MLX4_MCAST_CONFIG);
 904
 905                 /* Update multicast list - we cache all addresses so they won't
 906                  * change while HW is updated holding the command semaphor */
 907                 mlx4_en_cache_mclist(dev);
 908                 list_for_each_entry(mclist, &priv->mc_list, list) {
 909                         mcast_addr = mlx4_mac_to_u64(mclist->addr);
 910                         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port,
 911                                         mcast_addr, 0, MLX4_MCAST_CONFIG);
 912                 }
 913                 err = mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0,
 914                                           0, MLX4_MCAST_ENABLE);
 915                 if (err)
 916                         en_err(priv, "Failed enabling multicast filter\n");
 917
 918                 update_mclist_flags(priv, &priv->curr_list, &priv->mc_list);
 919                 list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
 920                         if (mclist->action == MCLIST_REM) {
 921                                 /* detach this address and delete from list */
 922                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
 923                                 mc_list[5] = priv->port;
 924                                 err = mlx4_multicast_detach(mdev->dev,
 925                                                             &priv->rss_map.indir_qp,
 926                                                             mc_list,
 927                                                             MLX4_PROT_ETH,
 928                                                             mclist->reg_id);
 929                                 if (err)
 930                                         en_err(priv, "Fail to detach multicast address\n");
 931
 932                                 /* remove from list */
 933                                 list_del(&mclist->list);
 934                                 kfree(mclist);
 935                         } else if (mclist->action == MCLIST_ADD) {
 936                                 /* attach the address */
 937                                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
 938                                 /* needed for B0 steering support */
 939                                 mc_list[5] = priv->port;
 940                                 err = mlx4_multicast_attach(mdev->dev,
 941                                                             &priv->rss_map.indir_qp,
 942                                                             mc_list,
 943                                                             priv->port, 0,
 944                                                             MLX4_PROT_ETH,
 945                                                             &mclist->reg_id);
 946                                 if (err)
 947                                         en_err(priv, "Fail to attach multicast address\n");
 948
 949                         }
 950                 }
 951         }
 952 }
 953
 954 static void mlx4_en_do_set_rx_mode(struct work_struct *work)
 955 {
 956         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
 957                                                  rx_mode_task);
 958         struct mlx4_en_dev *mdev = priv->mdev;
 959         struct net_device *dev = priv->dev;
 960
 961
 962         mutex_lock(&mdev->state_lock);
 963         if (!mdev->device_up) {
 964                 en_dbg(HW, priv, "Card is not up, ignoring rx mode change.\n");
 965                 goto out;
 966         }
 967         if (!priv->port_up) {
 968                 en_dbg(HW, priv, "Port is down, ignoring rx mode change.\n");
 969                 goto out;
 970         }
 971         if (!mlx4_en_QUERY_PORT(mdev, priv->port)) {
 972                 if (priv->port_state.link_state) {
 973                         priv->last_link_state = MLX4_DEV_EVENT_PORT_UP;
 974                         /* Important note: the following call for if_link_state_change
 975                          * is needed for interface up scenario (start port, link state
 976                          * change) */
 977                         /* update netif baudrate */
 978                         priv->dev->if_baudrate =
 979                             IF_Mbps(priv->port_state.link_speed);
 980                         if_link_state_change(priv->dev, LINK_STATE_UP);
 981                         en_dbg(HW, priv, "Link Up\n");
 982                 }
 983         }
 984
 985         /* Promsicuous mode: disable all filters */
 986         if ((dev->if_flags & IFF_PROMISC) ||
 987             (priv->flags & MLX4_EN_FLAG_FORCE_PROMISC)) {
 988                 mlx4_en_set_promisc_mode(priv, mdev);
 989                 goto out;
 990         }
 991
 992         /* Not in promiscuous mode */
 993         if (priv->flags & MLX4_EN_FLAG_PROMISC)
 994                 mlx4_en_clear_promisc_mode(priv, mdev);
 995
 996         mlx4_en_do_multicast(priv, dev, mdev);
 997 out:
 998         mutex_unlock(&mdev->state_lock);
 999 }
1000
1001 #ifdef CONFIG_NET_POLL_CONTROLLER
1002 static void mlx4_en_netpoll(struct net_device *dev)
1003 {
1004         struct mlx4_en_priv *priv = netdev_priv(dev);
1005         struct mlx4_en_cq *cq;
1006         unsigned long flags;
1007         int i;
1008
1009         for (i = 0; i < priv->rx_ring_num; i++) {
1010                 cq = priv->rx_cq[i];
1011                 spin_lock_irqsave(&cq->lock, flags);
1012                 napi_synchronize(&cq->napi);
1013                 mlx4_en_process_rx_cq(dev, cq, 0);
1014                 spin_unlock_irqrestore(&cq->lock, flags);
1015         }
1016 }
1017 #endif
1018
1019 static void mlx4_en_watchdog_timeout(void *arg)
1020 {
1021         struct mlx4_en_priv *priv = arg;
1022         struct mlx4_en_dev *mdev = priv->mdev;
1023
1024         en_dbg(DRV, priv, "Scheduling watchdog\n");
1025         queue_work(mdev->workqueue, &priv->watchdog_task);
1026         if (priv->port_up)
1027                 callout_reset(&priv->watchdog_timer, MLX4_EN_WATCHDOG_TIMEOUT,
1028                                 mlx4_en_watchdog_timeout, priv);
1029 }
1030
1031
1032
1033 static void mlx4_en_set_default_moderation(struct mlx4_en_priv *priv)
1034 {
1035         struct mlx4_en_cq *cq;
1036         int i;
1037
1038         /* If we haven't received a specific coalescing setting
1039          * (module param), we set the moderation parameters as follows:
1040          * - moder_cnt is set to the number of mtu sized packets to
1041          *   satisfy our coelsing target.
1042          * - moder_time is set to a fixed value.
1043          */
1044         priv->rx_frames = MLX4_EN_RX_COAL_TARGET / priv->dev->if_mtu + 1;
1045         priv->rx_usecs = MLX4_EN_RX_COAL_TIME;
1046         priv->tx_frames = MLX4_EN_TX_COAL_PKTS;
1047         priv->tx_usecs = MLX4_EN_TX_COAL_TIME;
1048         en_dbg(INTR, priv, "Default coalesing params for mtu: %u - "
1049                "rx_frames:%d rx_usecs:%d\n",
1050                (unsigned)priv->dev->if_mtu, priv->rx_frames, priv->rx_usecs);
1051
1052         /* Setup cq moderation params */
1053         for (i = 0; i < priv->rx_ring_num; i++) {
1054                 cq = priv->rx_cq[i];
1055                 cq->moder_cnt = priv->rx_frames;
1056                 cq->moder_time = priv->rx_usecs;
1057                 priv->last_moder_time[i] = MLX4_EN_AUTO_CONF;
1058                 priv->last_moder_packets[i] = 0;
1059                 priv->last_moder_bytes[i] = 0;
1060         }
1061
1062         for (i = 0; i < priv->tx_ring_num; i++) {
1063                 cq = priv->tx_cq[i];
1064                 cq->moder_cnt = priv->tx_frames;
1065                 cq->moder_time = priv->tx_usecs;
1066         }
1067
1068         /* Reset auto-moderation params */
1069         priv->pkt_rate_low = MLX4_EN_RX_RATE_LOW;
1070         priv->rx_usecs_low = MLX4_EN_RX_COAL_TIME_LOW;
1071         priv->pkt_rate_high = MLX4_EN_RX_RATE_HIGH;
1072         priv->rx_usecs_high = MLX4_EN_RX_COAL_TIME_HIGH;
1073         priv->sample_interval = MLX4_EN_SAMPLE_INTERVAL;
1074         priv->adaptive_rx_coal = 1;
1075         priv->last_moder_jiffies = 0;
1076         priv->last_moder_tx_packets = 0;
1077 }
1078
1079 static void mlx4_en_auto_moderation(struct mlx4_en_priv *priv)
1080 {
1081         unsigned long period = (unsigned long) (jiffies - priv->last_moder_jiffies);
1082         struct mlx4_en_cq *cq;
1083         unsigned long packets;
1084         unsigned long rate;
1085         unsigned long avg_pkt_size;
1086         unsigned long rx_packets;
1087         unsigned long rx_bytes;
1088         unsigned long rx_pkt_diff;
1089         int moder_time;
1090         int ring, err;
1091
1092         if (!priv->adaptive_rx_coal || period < priv->sample_interval * HZ)
1093                 return;
1094
1095         for (ring = 0; ring < priv->rx_ring_num; ring++) {
1096                 spin_lock(&priv->stats_lock);
1097                 rx_packets = priv->rx_ring[ring]->packets;
1098                 rx_bytes = priv->rx_ring[ring]->bytes;
1099                 spin_unlock(&priv->stats_lock);
1100
1101                 rx_pkt_diff = ((unsigned long) (rx_packets -
1102                                 priv->last_moder_packets[ring]));
1103                 packets = rx_pkt_diff;
1104                 rate = packets * HZ / period;
1105                 avg_pkt_size = packets ? ((unsigned long) (rx_bytes -
1106                                 priv->last_moder_bytes[ring])) / packets : 0;
1107
1108                 /* Apply auto-moderation only when packet rate
1109                  * exceeds a rate that it matters */
1110                 if (rate > (MLX4_EN_RX_RATE_THRESH / priv->rx_ring_num) &&
1111                     avg_pkt_size > MLX4_EN_AVG_PKT_SMALL) {
1112                         if (rate < priv->pkt_rate_low)
1113                                 moder_time = priv->rx_usecs_low;
1114                         else if (rate > priv->pkt_rate_high)
1115                                 moder_time = priv->rx_usecs_high;
1116                         else
1117                                 moder_time = (rate - priv->pkt_rate_low) *
1118                                         (priv->rx_usecs_high - priv->rx_usecs_low) /
1119                                         (priv->pkt_rate_high - priv->pkt_rate_low) +
1120                                         priv->rx_usecs_low;
1121                 } else {
1122                         moder_time = priv->rx_usecs_low;
1123                 }
1124
1125                 if (moder_time != priv->last_moder_time[ring]) {
1126                         priv->last_moder_time[ring] = moder_time;
1127                         cq = priv->rx_cq[ring];
1128                         cq->moder_time = moder_time;
1129                         err = mlx4_en_set_cq_moder(priv, cq);
1130                         if (err)
1131                                 en_err(priv, "Failed modifying moderation for cq:%d\n",
1132                                        ring);
1133                 }
1134                 priv->last_moder_packets[ring] = rx_packets;
1135                 priv->last_moder_bytes[ring] = rx_bytes;
1136         }
1137
1138         priv->last_moder_jiffies = jiffies;
1139 }
1140
1141 static void mlx4_en_do_get_stats(struct work_struct *work)
1142 {
1143         struct delayed_work *delay = to_delayed_work(work);
1144         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1145                                                  stats_task);
1146         struct mlx4_en_dev *mdev = priv->mdev;
1147         int err;
1148
1149         mutex_lock(&mdev->state_lock);
1150         if (mdev->device_up) {
1151                 if (priv->port_up) {
1152                         err = mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 0);
1153                         if (err)
1154                                 en_dbg(HW, priv, "Could not update stats\n");
1155
1156                         mlx4_en_auto_moderation(priv);
1157                 }
1158
1159                 queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
1160         }
1161         mutex_unlock(&mdev->state_lock);
1162 }
1163
1164 /* mlx4_en_service_task - Run service task for tasks that needed to be done
1165  * periodically
1166  */
1167 static void mlx4_en_service_task(struct work_struct *work)
1168 {
1169         struct delayed_work *delay = to_delayed_work(work);
1170         struct mlx4_en_priv *priv = container_of(delay, struct mlx4_en_priv,
1171                                                  service_task);
1172         struct mlx4_en_dev *mdev = priv->mdev;
1173
1174         mutex_lock(&mdev->state_lock);
1175         if (mdev->device_up) {
1176                 queue_delayed_work(mdev->workqueue, &priv->service_task,
1177                                    SERVICE_TASK_DELAY);
1178         }
1179         mutex_unlock(&mdev->state_lock);
1180 }
1181
1182 static void mlx4_en_linkstate(struct work_struct *work)
1183 {
1184         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1185                                                  linkstate_task);
1186         struct mlx4_en_dev *mdev = priv->mdev;
1187         int linkstate = priv->link_state;
1188
1189         mutex_lock(&mdev->state_lock);
1190         /* If observable port state changed set carrier state and
1191          * report to system log */
1192         if (priv->last_link_state != linkstate) {
1193                 if (linkstate == MLX4_DEV_EVENT_PORT_DOWN) {
1194                         en_info(priv, "Link Down\n");
1195                         if_link_state_change(priv->dev, LINK_STATE_DOWN);
1196                         /* update netif baudrate */
1197                         priv->dev->if_baudrate = 0;
1198
1199                 /* make sure the port is up before notifying the OS.
1200                  * This is tricky since we get here on INIT_PORT and
1201                  * in such case we can't tell the OS the port is up.
1202                  * To solve this there is a call to if_link_state_change
1203                  * in set_rx_mode.
1204                  * */
1205                 } else if (priv->port_up && (linkstate == MLX4_DEV_EVENT_PORT_UP)){
1206                         if (mlx4_en_QUERY_PORT(priv->mdev, priv->port))
1207                                 en_info(priv, "Query port failed\n");
1208                         priv->dev->if_baudrate =
1209                             IF_Mbps(priv->port_state.link_speed);
1210                         en_info(priv, "Link Up\n");
1211                         if_link_state_change(priv->dev, LINK_STATE_UP);
1212                 }
1213         }
1214         priv->last_link_state = linkstate;
1215         mutex_unlock(&mdev->state_lock);
1216 }
1217
1218
1219 int mlx4_en_start_port(struct net_device *dev)
1220 {
1221         struct mlx4_en_priv *priv = netdev_priv(dev);
1222         struct mlx4_en_dev *mdev = priv->mdev;
1223         struct mlx4_en_cq *cq;
1224         struct mlx4_en_tx_ring *tx_ring;
1225         int rx_index = 0;
1226         int tx_index = 0;
1227         int err = 0;
1228         int i;
1229         int j;
1230         u8 mc_list[16] = {0};
1231
1232
1233         if (priv->port_up) {
1234                 en_dbg(DRV, priv, "start port called while port already up\n");
1235                 return 0;
1236         }
1237
1238         INIT_LIST_HEAD(&priv->mc_list);
1239         INIT_LIST_HEAD(&priv->curr_list);
1240         INIT_LIST_HEAD(&priv->ethtool_list);
1241
1242         /* Calculate Rx buf size */
1243         dev->if_mtu = min(dev->if_mtu, priv->max_mtu);
1244         mlx4_en_calc_rx_buf(dev);
1245         priv->rx_alloc_size = max_t(int, 2 * roundup_pow_of_two(priv->rx_mb_size),
1246                                     PAGE_SIZE);
1247         priv->rx_alloc_order = get_order(priv->rx_alloc_size);
1248         priv->rx_buf_size = roundup_pow_of_two(priv->rx_mb_size);
1249         priv->log_rx_info = ROUNDUP_LOG2(sizeof(struct mlx4_en_rx_buf));
1250         en_dbg(DRV, priv, "Rx buf size:%d\n", priv->rx_mb_size);
1251
1252         /* Configure rx cq's and rings */
1253         err = mlx4_en_activate_rx_rings(priv);
1254         if (err) {
1255                 en_err(priv, "Failed to activate RX rings\n");
1256                 return err;
1257         }
1258         for (i = 0; i < priv->rx_ring_num; i++) {
1259                 cq = priv->rx_cq[i];
1260
1261                 mlx4_en_cq_init_lock(cq);
1262                 err = mlx4_en_activate_cq(priv, cq, i);
1263                 if (err) {
1264                         en_err(priv, "Failed activating Rx CQ\n");
1265                         goto cq_err;
1266                 }
1267                 for (j = 0; j < cq->size; j++)
1268                         cq->buf[j].owner_sr_opcode = MLX4_CQE_OWNER_MASK;
1269                 err = mlx4_en_set_cq_moder(priv, cq);
1270                 if (err) {
1271                         en_err(priv, "Failed setting cq moderation parameters");
1272                         mlx4_en_deactivate_cq(priv, cq);
1273                         goto cq_err;
1274                 }
1275                 mlx4_en_arm_cq(priv, cq);
1276                 priv->rx_ring[i]->cqn = cq->mcq.cqn;
1277                 ++rx_index;
1278         }
1279
1280         /* Set qp number */
1281         en_dbg(DRV, priv, "Getting qp number for port %d\n", priv->port);
1282         err = mlx4_en_get_qp(priv);
1283         if (err) {
1284                 en_err(priv, "Failed getting eth qp\n");
1285                 goto cq_err;
1286         }
1287         mdev->mac_removed[priv->port] = 0;
1288
1289         /* gets default allocated counter index from func cap */
1290         /* or sink counter index if no resources */
1291         priv->counter_index = mdev->dev->caps.def_counter_index[priv->port - 1];
1292
1293         en_dbg(DRV, priv, "%s: default counter index %d for port %d\n",
1294                __func__, priv->counter_index, priv->port);
1295
1296         err = mlx4_en_config_rss_steer(priv);
1297         if (err) {
1298                 en_err(priv, "Failed configuring rss steering\n");
1299                 goto mac_err;
1300         }
1301
1302         err = mlx4_en_create_drop_qp(priv);
1303         if (err)
1304                 goto rss_err;
1305
1306         /* Configure tx cq's and rings */
1307         for (i = 0; i < priv->tx_ring_num; i++) {
1308                 /* Configure cq */
1309                 cq = priv->tx_cq[i];
1310                 err = mlx4_en_activate_cq(priv, cq, i);
1311                 if (err) {
1312                         en_err(priv, "Failed activating Tx CQ\n");
1313                         goto tx_err;
1314                 }
1315                 err = mlx4_en_set_cq_moder(priv, cq);
1316                 if (err) {
1317                         en_err(priv, "Failed setting cq moderation parameters");
1318                         mlx4_en_deactivate_cq(priv, cq);
1319                         goto tx_err;
1320                 }
1321                 en_dbg(DRV, priv, "Resetting index of collapsed CQ:%d to -1\n", i);
1322                 cq->buf->wqe_index = cpu_to_be16(0xffff);
1323
1324                 /* Configure ring */
1325                 tx_ring = priv->tx_ring[i];
1326
1327                 err = mlx4_en_activate_tx_ring(priv, tx_ring, cq->mcq.cqn,
1328                                                i / priv->num_tx_rings_p_up);
1329                 if (err) {
1330                         en_err(priv, "Failed activating Tx ring %d\n", i);
1331                         mlx4_en_deactivate_cq(priv, cq);
1332                         goto tx_err;
1333                 }
1334
1335                 /* Arm CQ for TX completions */
1336                 mlx4_en_arm_cq(priv, cq);
1337
1338                 /* Set initial ownership of all Tx TXBBs to SW (1) */
1339                 for (j = 0; j < tx_ring->buf_size; j += STAMP_STRIDE)
1340                         *((u32 *) (tx_ring->buf + j)) = 0xffffffff;
1341                 ++tx_index;
1342         }
1343
1344         /* Configure port */
1345         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
1346                                     priv->rx_mb_size,
1347                                     priv->prof->tx_pause,
1348                                     priv->prof->tx_ppp,
1349                                     priv->prof->rx_pause,
1350                                     priv->prof->rx_ppp);
1351         if (err) {
1352                 en_err(priv, "Failed setting port general configurations for port %d, with error %d\n",
1353                        priv->port, err);
1354                 goto tx_err;
1355         }
1356         /* Set default qp number */
1357         err = mlx4_SET_PORT_qpn_calc(mdev->dev, priv->port, priv->base_qpn, 0);
1358         if (err) {
1359                 en_err(priv, "Failed setting default qp numbers\n");
1360                 goto tx_err;
1361         }
1362
1363         /* Init port */
1364         en_dbg(HW, priv, "Initializing port\n");
1365         err = mlx4_INIT_PORT(mdev->dev, priv->port);
1366         if (err) {
1367                 en_err(priv, "Failed Initializing port\n");
1368                 goto tx_err;
1369         }
1370
1371         /* Attach rx QP to bradcast address */
1372         memset(&mc_list[10], 0xff, ETH_ALEN);
1373         mc_list[5] = priv->port; /* needed for B0 steering support */
1374         if (mlx4_multicast_attach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1375                                   priv->port, 0, MLX4_PROT_ETH,
1376                                   &priv->broadcast_id))
1377                 mlx4_warn(mdev, "Failed Attaching Broadcast\n");
1378
1379         /* Must redo promiscuous mode setup. */
1380         priv->flags &= ~(MLX4_EN_FLAG_PROMISC | MLX4_EN_FLAG_MC_PROMISC);
1381
1382         /* Schedule multicast task to populate multicast list */
1383         queue_work(mdev->workqueue, &priv->rx_mode_task);
1384
1385         mlx4_set_stats_bitmap(mdev->dev, priv->stats_bitmap);
1386
1387         priv->port_up = true;
1388
1389         /* Enable the queues. */
1390         dev->if_drv_flags &= ~IFF_DRV_OACTIVE;
1391         dev->if_drv_flags |= IFF_DRV_RUNNING;
1392 #ifdef CONFIG_DEBUG_FS
1393         mlx4_en_create_debug_files(priv);
1394 #endif
1395         callout_reset(&priv->watchdog_timer, MLX4_EN_WATCHDOG_TIMEOUT,
1396                     mlx4_en_watchdog_timeout, priv);
1397
1398
1399         return 0;
1400
1401 tx_err:
1402         while (tx_index--) {
1403                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[tx_index]);
1404                 mlx4_en_deactivate_cq(priv, priv->tx_cq[tx_index]);
1405         }
1406         mlx4_en_destroy_drop_qp(priv);
1407 rss_err:
1408         mlx4_en_release_rss_steer(priv);
1409 mac_err:
1410         mlx4_en_put_qp(priv);
1411 cq_err:
1412         while (rx_index--)
1413                 mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
1414         for (i = 0; i < priv->rx_ring_num; i++)
1415                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1416
1417         return err; /* need to close devices */
1418 }
1419
1420
1421 void mlx4_en_stop_port(struct net_device *dev)
1422 {
1423         struct mlx4_en_priv *priv = netdev_priv(dev);
1424         struct mlx4_en_dev *mdev = priv->mdev;
1425         struct mlx4_en_mc_list *mclist, *tmp;
1426         int i;
1427         u8 mc_list[16] = {0};
1428
1429         if (!priv->port_up) {
1430                 en_dbg(DRV, priv, "stop port called while port already down\n");
1431                 return;
1432         }
1433
1434 #ifdef CONFIG_DEBUG_FS
1435         mlx4_en_delete_debug_files(priv);
1436 #endif
1437
1438         /* close port*/
1439         mlx4_CLOSE_PORT(mdev->dev, priv->port);
1440
1441         /* Set port as not active */
1442         priv->port_up = false;
1443         if (priv->counter_index != 0xff) {
1444                 mlx4_counter_free(mdev->dev, priv->port, priv->counter_index);
1445                 priv->counter_index = 0xff;
1446         }
1447
1448         /* Promsicuous mode */
1449         if (mdev->dev->caps.steering_mode ==
1450             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1451                 priv->flags &= ~(MLX4_EN_FLAG_PROMISC |
1452                                  MLX4_EN_FLAG_MC_PROMISC);
1453                 mlx4_flow_steer_promisc_remove(mdev->dev,
1454                                                priv->port,
1455                                                MLX4_FS_ALL_DEFAULT);
1456                 mlx4_flow_steer_promisc_remove(mdev->dev,
1457                                                priv->port,
1458                                                MLX4_FS_MC_DEFAULT);
1459         } else if (priv->flags & MLX4_EN_FLAG_PROMISC) {
1460                 priv->flags &= ~MLX4_EN_FLAG_PROMISC;
1461
1462                 /* Disable promiscouos mode */
1463                 mlx4_unicast_promisc_remove(mdev->dev, priv->base_qpn,
1464                                             priv->port);
1465
1466                 /* Disable Multicast promisc */
1467                 if (priv->flags & MLX4_EN_FLAG_MC_PROMISC) {
1468                         mlx4_multicast_promisc_remove(mdev->dev, priv->base_qpn,
1469                                                       priv->port);
1470                         priv->flags &= ~MLX4_EN_FLAG_MC_PROMISC;
1471                 }
1472         }
1473
1474         /* Detach All multicasts */
1475         memset(&mc_list[10], 0xff, ETH_ALEN);
1476         mc_list[5] = priv->port; /* needed for B0 steering support */
1477         mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp, mc_list,
1478                               MLX4_PROT_ETH, priv->broadcast_id);
1479         list_for_each_entry(mclist, &priv->curr_list, list) {
1480                 memcpy(&mc_list[10], mclist->addr, ETH_ALEN);
1481                 mc_list[5] = priv->port;
1482                 mlx4_multicast_detach(mdev->dev, &priv->rss_map.indir_qp,
1483                                       mc_list, MLX4_PROT_ETH, mclist->reg_id);
1484         }
1485         mlx4_en_clear_list(dev);
1486         list_for_each_entry_safe(mclist, tmp, &priv->curr_list, list) {
1487                 list_del(&mclist->list);
1488                 kfree(mclist);
1489         }
1490
1491         /* Flush multicast filter */
1492         mlx4_SET_MCAST_FLTR(mdev->dev, priv->port, 0, 1, MLX4_MCAST_CONFIG);
1493         mlx4_en_destroy_drop_qp(priv);
1494
1495         /* Free TX Rings */
1496         for (i = 0; i < priv->tx_ring_num; i++) {
1497                 mlx4_en_deactivate_tx_ring(priv, priv->tx_ring[i]);
1498                 mlx4_en_deactivate_cq(priv, priv->tx_cq[i]);
1499         }
1500         msleep(10);
1501
1502         for (i = 0; i < priv->tx_ring_num; i++)
1503                 mlx4_en_free_tx_buf(dev, priv->tx_ring[i]);
1504
1505         /* Free RSS qps */
1506         mlx4_en_release_rss_steer(priv);
1507
1508         /* Unregister Mac address for the port */
1509         mlx4_en_put_qp(priv);
1510         mdev->mac_removed[priv->port] = 1;
1511
1512         /* Free RX Rings */
1513         for (i = 0; i < priv->rx_ring_num; i++) {
1514                 struct mlx4_en_cq *cq = priv->rx_cq[i];
1515                 mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
1516                 mlx4_en_deactivate_cq(priv, cq);
1517         }
1518
1519         callout_stop(&priv->watchdog_timer);
1520
1521         dev->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1522 }
1523
1524 static void mlx4_en_restart(struct work_struct *work)
1525 {
1526         struct mlx4_en_priv *priv = container_of(work, struct mlx4_en_priv,
1527                                                  watchdog_task);
1528         struct mlx4_en_dev *mdev = priv->mdev;
1529         struct net_device *dev = priv->dev;
1530         struct mlx4_en_tx_ring *ring;
1531         int i;
1532
1533
1534         if (priv->blocked == 0 || priv->port_up == 0)
1535                 return;
1536         for (i = 0; i < priv->tx_ring_num; i++) {
1537                 ring = priv->tx_ring[i];
1538                 if (ring->blocked &&
1539                                 ring->watchdog_time + MLX4_EN_WATCHDOG_TIMEOUT < ticks)
1540                         goto reset;
1541         }
1542         return;
1543
1544 reset:
1545         priv->port_stats.tx_timeout++;
1546         en_dbg(DRV, priv, "Watchdog task called for port %d\n", priv->port);
1547
1548         mutex_lock(&mdev->state_lock);
1549         if (priv->port_up) {
1550                 mlx4_en_stop_port(dev);
1551                 //for (i = 0; i < priv->tx_ring_num; i++)
1552                 //        netdev_tx_reset_queue(priv->tx_ring[i]->tx_queue);
1553                 if (mlx4_en_start_port(dev))
1554                         en_err(priv, "Failed restarting port %d\n", priv->port);
1555         }
1556         mutex_unlock(&mdev->state_lock);
1557 }
1558
1559 static void mlx4_en_clear_stats(struct net_device *dev)
1560 {
1561         struct mlx4_en_priv *priv = netdev_priv(dev);
1562         struct mlx4_en_dev *mdev = priv->mdev;
1563         int i;
1564
1565         if (!mlx4_is_slave(mdev->dev))
1566                 if (mlx4_en_DUMP_ETH_STATS(mdev, priv->port, 1))
1567                         en_dbg(HW, priv, "Failed dumping statistics\n");
1568
1569         memset(&priv->pstats, 0, sizeof(priv->pstats));
1570         memset(&priv->pkstats, 0, sizeof(priv->pkstats));
1571         memset(&priv->port_stats, 0, sizeof(priv->port_stats));
1572         memset(&priv->vport_stats, 0, sizeof(priv->vport_stats));
1573
1574         for (i = 0; i < priv->tx_ring_num; i++) {
1575                 priv->tx_ring[i]->bytes = 0;
1576                 priv->tx_ring[i]->packets = 0;
1577                 priv->tx_ring[i]->tx_csum = 0;
1578         }
1579         for (i = 0; i < priv->rx_ring_num; i++) {
1580                 priv->rx_ring[i]->bytes = 0;
1581                 priv->rx_ring[i]->packets = 0;
1582                 priv->rx_ring[i]->csum_ok = 0;
1583                 priv->rx_ring[i]->csum_none = 0;
1584         }
1585 }
1586
1587 static void mlx4_en_open(void* arg)
1588 {
1589
1590         struct mlx4_en_priv *priv;
1591         struct mlx4_en_dev *mdev;
1592         struct net_device *dev;
1593         int err = 0;
1594
1595         priv = arg;
1596         mdev = priv->mdev;
1597         dev = priv->dev;
1598
1599
1600         mutex_lock(&mdev->state_lock);
1601
1602         if (!mdev->device_up) {
1603                 en_err(priv, "Cannot open - device down/disabled\n");
1604                 goto out;
1605         }
1606
1607         /* Reset HW statistics and SW counters */
1608         mlx4_en_clear_stats(dev);
1609
1610         err = mlx4_en_start_port(dev);
1611         if (err)
1612                 en_err(priv, "Failed starting port:%d\n", priv->port);
1613
1614 out:
1615         mutex_unlock(&mdev->state_lock);
1616         return;
1617 }
1618
1619 void mlx4_en_free_resources(struct mlx4_en_priv *priv)
1620 {
1621         int i;
1622
1623 #ifdef CONFIG_RFS_ACCEL
1624         if (priv->dev->rx_cpu_rmap) {
1625                 free_irq_cpu_rmap(priv->dev->rx_cpu_rmap);
1626                 priv->dev->rx_cpu_rmap = NULL;
1627         }
1628 #endif
1629
1630         for (i = 0; i < priv->tx_ring_num; i++) {
1631                 if (priv->tx_ring && priv->tx_ring[i])
1632                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1633                 if (priv->tx_cq && priv->tx_cq[i])
1634                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1635         }
1636
1637         for (i = 0; i < priv->rx_ring_num; i++) {
1638                 if (priv->rx_ring[i])
1639                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1640                                 priv->prof->rx_ring_size, priv->stride);
1641                 if (priv->rx_cq[i])
1642                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1643         }
1644
1645         if (priv->sysctl)
1646                 sysctl_ctx_free(&priv->stat_ctx);
1647
1648
1649 }
1650
1651 int mlx4_en_alloc_resources(struct mlx4_en_priv *priv)
1652 {
1653         struct mlx4_en_port_profile *prof = priv->prof;
1654         int i;
1655         int node = 0;
1656
1657         /* Create rx Rings */
1658         for (i = 0; i < priv->rx_ring_num; i++) {
1659                 if (mlx4_en_create_cq(priv, &priv->rx_cq[i],
1660                                       prof->rx_ring_size, i, RX, node))
1661                         goto err;
1662
1663                 if (mlx4_en_create_rx_ring(priv, &priv->rx_ring[i],
1664                                            prof->rx_ring_size, node))
1665                         goto err;
1666         }
1667
1668         /* Create tx Rings */
1669         for (i = 0; i < priv->tx_ring_num; i++) {
1670                 if (mlx4_en_create_cq(priv, &priv->tx_cq[i],
1671                                       prof->tx_ring_size, i, TX, node))
1672                         goto err;
1673
1674                 if (mlx4_en_create_tx_ring(priv, &priv->tx_ring[i],
1675                                            prof->tx_ring_size, TXBB_SIZE, node, i))
1676                         goto err;
1677         }
1678
1679 #ifdef CONFIG_RFS_ACCEL
1680         priv->dev->rx_cpu_rmap = alloc_irq_cpu_rmap(priv->rx_ring_num);
1681         if (!priv->dev->rx_cpu_rmap)
1682                 goto err;
1683 #endif
1684         /* Re-create stat sysctls in case the number of rings changed. */
1685         mlx4_en_sysctl_stat(priv);
1686         return 0;
1687
1688 err:
1689         en_err(priv, "Failed to allocate NIC resources\n");
1690         for (i = 0; i < priv->rx_ring_num; i++) {
1691                 if (priv->rx_ring[i])
1692                         mlx4_en_destroy_rx_ring(priv, &priv->rx_ring[i],
1693                                                 prof->rx_ring_size,
1694                                                 priv->stride);
1695                 if (priv->rx_cq[i])
1696                         mlx4_en_destroy_cq(priv, &priv->rx_cq[i]);
1697         }
1698         for (i = 0; i < priv->tx_ring_num; i++) {
1699                 if (priv->tx_ring[i])
1700                         mlx4_en_destroy_tx_ring(priv, &priv->tx_ring[i]);
1701                 if (priv->tx_cq[i])
1702                         mlx4_en_destroy_cq(priv, &priv->tx_cq[i]);
1703         }
1704         priv->port_up = false;
1705         return -ENOMEM;
1706 }
1707
1708 struct en_port_attribute {
1709         struct attribute attr;
1710         ssize_t (*show)(struct en_port *, struct en_port_attribute *, char *buf);
1711         ssize_t (*store)(struct en_port *, struct en_port_attribute *, char *buf, size_t count);
1712 };
1713
1714 #define PORT_ATTR_RO(_name) \
1715 struct en_port_attribute en_port_attr_##_name = __ATTR_RO(_name)
1716
1717 #define EN_PORT_ATTR(_name, _mode, _show, _store) \
1718 struct en_port_attribute en_port_attr_##_name = __ATTR(_name, _mode, _show, _store)
1719
1720 void mlx4_en_destroy_netdev(struct net_device *dev)
1721 {
1722         struct mlx4_en_priv *priv = netdev_priv(dev);
1723         struct mlx4_en_dev *mdev = priv->mdev;
1724
1725         en_dbg(DRV, priv, "Destroying netdev on port:%d\n", priv->port);
1726
1727         if (priv->vlan_attach != NULL)
1728                 EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach);
1729         if (priv->vlan_detach != NULL)
1730                 EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach);
1731
1732         /* Unregister device - this will close the port if it was up */
1733         if (priv->registered)
1734                 ether_ifdetach(dev);
1735
1736         if (priv->allocated)
1737                 mlx4_free_hwq_res(mdev->dev, &priv->res, MLX4_EN_PAGE_SIZE);
1738
1739         mutex_lock(&mdev->state_lock);
1740         mlx4_en_stop_port(dev);
1741         mutex_unlock(&mdev->state_lock);
1742
1743
1744         cancel_delayed_work(&priv->stats_task);
1745         cancel_delayed_work(&priv->service_task);
1746         /* flush any pending task for this netdev */
1747         flush_workqueue(mdev->workqueue);
1748         callout_drain(&priv->watchdog_timer);
1749
1750         /* Detach the netdev so tasks would not attempt to access it */
1751         mutex_lock(&mdev->state_lock);
1752         mdev->pndev[priv->port] = NULL;
1753         mutex_unlock(&mdev->state_lock);
1754
1755
1756         mlx4_en_free_resources(priv);
1757
1758         /* freeing the sysctl conf cannot be called from within mlx4_en_free_resources */
1759         if (priv->sysctl)
1760                 sysctl_ctx_free(&priv->conf_ctx);
1761
1762         kfree(priv->tx_ring);
1763         kfree(priv->tx_cq);
1764
1765         kfree(priv);
1766         if_free(dev);
1767
1768 }
1769
1770 static int mlx4_en_change_mtu(struct net_device *dev, int new_mtu)
1771 {
1772         struct mlx4_en_priv *priv = netdev_priv(dev);
1773         struct mlx4_en_dev *mdev = priv->mdev;
1774         int err = 0;
1775
1776         en_dbg(DRV, priv, "Change MTU called - current:%u new:%u\n",
1777                (unsigned)dev->if_mtu, (unsigned)new_mtu);
1778
1779         if ((new_mtu < MLX4_EN_MIN_MTU) || (new_mtu > priv->max_mtu)) {
1780                 en_err(priv, "Bad MTU size:%d.\n", new_mtu);
1781                 return -EPERM;
1782         }
1783         mutex_lock(&mdev->state_lock);
1784         dev->if_mtu = new_mtu;
1785         if (dev->if_drv_flags & IFF_DRV_RUNNING) {
1786                 if (!mdev->device_up) {
1787                         /* NIC is probably restarting - let watchdog task reset
1788                          *                          * the port */
1789                         en_dbg(DRV, priv, "Change MTU called with card down!?\n");
1790                 } else {
1791                         mlx4_en_stop_port(dev);
1792                         err = mlx4_en_start_port(dev);
1793                         if (err) {
1794                                 en_err(priv, "Failed restarting port:%d\n",
1795                                                 priv->port);
1796                                 queue_work(mdev->workqueue, &priv->watchdog_task);
1797                         }
1798                 }
1799         }
1800         mutex_unlock(&mdev->state_lock);
1801         return 0;
1802 }
1803
1804 static int mlx4_en_calc_media(struct mlx4_en_priv *priv)
1805 {
1806         int trans_type;
1807         int active;
1808
1809         active = IFM_ETHER;
1810         if (priv->last_link_state == MLX4_DEV_EVENT_PORT_DOWN)
1811                 return (active);
1812         /*
1813          * [ShaharK] mlx4_en_QUERY_PORT sleeps and cannot be called under a
1814          * non-sleepable lock.
1815          * I moved it to the periodic mlx4_en_do_get_stats.
1816         if (mlx4_en_QUERY_PORT(priv->mdev, priv->port))
1817                 return (active);
1818         */
1819         active |= IFM_FDX;
1820         trans_type = priv->port_state.transciver;
1821         /* XXX I don't know all of the transceiver values. */
1822         switch (priv->port_state.link_speed) {
1823         case 1000:
1824                 active |= IFM_1000_T;
1825                 break;
1826         case 10000:
1827                 if (trans_type > 0 && trans_type <= 0xC)
1828                         active |= IFM_10G_SR;
1829                 else if (trans_type == 0x80 || trans_type == 0)
1830                         active |= IFM_10G_CX4;
1831                 break;
1832         case 40000:
1833                 active |= IFM_40G_CR4;
1834                 break;
1835         }
1836         if (priv->prof->tx_pause)
1837                 active |= IFM_ETH_TXPAUSE;
1838         if (priv->prof->rx_pause)
1839                 active |= IFM_ETH_RXPAUSE;
1840
1841         return (active);
1842 }
1843
1844 static void mlx4_en_media_status(struct ifnet *dev, struct ifmediareq *ifmr)
1845 {
1846         struct mlx4_en_priv *priv;
1847
1848         priv = dev->if_softc;
1849         ifmr->ifm_status = IFM_AVALID;
1850         if (priv->last_link_state != MLX4_DEV_EVENT_PORT_DOWN)
1851                 ifmr->ifm_status |= IFM_ACTIVE;
1852         ifmr->ifm_active = mlx4_en_calc_media(priv);
1853
1854         return;
1855 }
1856
1857 static int mlx4_en_media_change(struct ifnet *dev)
1858 {
1859         struct mlx4_en_priv *priv;
1860         struct ifmedia *ifm;
1861         int rxpause;
1862         int txpause;
1863         int error;
1864
1865         priv = dev->if_softc;
1866         ifm = &priv->media;
1867         rxpause = txpause = 0;
1868         error = 0;
1869
1870         if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1871                 return (EINVAL);
1872         switch (IFM_SUBTYPE(ifm->ifm_media)) {
1873         case IFM_AUTO:
1874                 break;
1875         case IFM_10G_SR:
1876         case IFM_10G_CX4:
1877         case IFM_1000_T:
1878         case IFM_40G_CR4:
1879                 if ((IFM_SUBTYPE(ifm->ifm_media)
1880                         == IFM_SUBTYPE(mlx4_en_calc_media(priv)))
1881                         && (ifm->ifm_media & IFM_FDX))
1882                         break;
1883                 /* Fallthrough */
1884         default:
1885                 printf("%s: Only auto media type\n", if_name(dev));
1886                 return (EINVAL);
1887         }
1888         /* Allow user to set/clear pause */
1889         if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
1890                 rxpause = 1;
1891         if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
1892                 txpause = 1;
1893         if (priv->prof->tx_pause != txpause || priv->prof->rx_pause != rxpause) {
1894                 priv->prof->tx_pause = txpause;
1895                 priv->prof->rx_pause = rxpause;
1896                 error = -mlx4_SET_PORT_general(priv->mdev->dev, priv->port,
1897                      priv->rx_mb_size + ETHER_CRC_LEN, priv->prof->tx_pause,
1898                      priv->prof->tx_ppp, priv->prof->rx_pause,
1899                      priv->prof->rx_ppp);
1900         }
1901         return (error);
1902 }
1903
1904 static int mlx4_en_ioctl(struct ifnet *dev, u_long command, caddr_t data)
1905 {
1906         struct mlx4_en_priv *priv;
1907         struct mlx4_en_dev *mdev;
1908         struct ifreq *ifr;
1909         int error;
1910         int mask;
1911
1912         error = 0;
1913         mask = 0;
1914         priv = dev->if_softc;
1915         mdev = priv->mdev;
1916         ifr = (struct ifreq *) data;
1917         switch (command) {
1918
1919         case SIOCSIFMTU:
1920                 error = -mlx4_en_change_mtu(dev, ifr->ifr_mtu);
1921                 break;
1922         case SIOCSIFFLAGS:
1923                 if (dev->if_flags & IFF_UP) {
1924                         if ((dev->if_drv_flags & IFF_DRV_RUNNING) == 0) {
1925                                 mutex_lock(&mdev->state_lock);
1926                                 mlx4_en_start_port(dev);
1927                                 mutex_unlock(&mdev->state_lock);
1928                         } else {
1929                                 mlx4_en_set_rx_mode(dev);
1930                         }
1931                 } else {
1932                         mutex_lock(&mdev->state_lock);
1933                         if (dev->if_drv_flags & IFF_DRV_RUNNING) {
1934                                 mlx4_en_stop_port(dev);
1935                                 if_link_state_change(dev, LINK_STATE_DOWN);
1936                         }
1937                         mutex_unlock(&mdev->state_lock);
1938                 }
1939                 break;
1940         case SIOCADDMULTI:
1941         case SIOCDELMULTI:
1942                 mlx4_en_set_rx_mode(dev);
1943                 break;
1944         case SIOCSIFMEDIA:
1945         case SIOCGIFMEDIA:
1946                 error = ifmedia_ioctl(dev, ifr, &priv->media, command);
1947                 break;
1948         case SIOCSIFCAP:
1949                 mutex_lock(&mdev->state_lock);
1950                 mask = ifr->ifr_reqcap ^ dev->if_capenable;
1951                 if (mask & IFCAP_HWCSUM)
1952                         dev->if_capenable ^= IFCAP_HWCSUM;
1953                 if (mask & IFCAP_TSO4)
1954                         dev->if_capenable ^= IFCAP_TSO4;
1955                 if (mask & IFCAP_TSO6)
1956                         dev->if_capenable ^= IFCAP_TSO6;
1957                 if (mask & IFCAP_LRO)
1958                         dev->if_capenable ^= IFCAP_LRO;
1959                 if (mask & IFCAP_VLAN_HWTAGGING)
1960                         dev->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1961                 if (mask & IFCAP_VLAN_HWFILTER)
1962                         dev->if_capenable ^= IFCAP_VLAN_HWFILTER;
1963                 if (mask & IFCAP_WOL_MAGIC)
1964                         dev->if_capenable ^= IFCAP_WOL_MAGIC;
1965                 if (dev->if_drv_flags & IFF_DRV_RUNNING)
1966                         mlx4_en_start_port(dev);
1967                 mutex_unlock(&mdev->state_lock);
1968                 VLAN_CAPABILITIES(dev);
1969                 break;
1970         default:
1971                 error = ether_ioctl(dev, command, data);
1972                 break;
1973         }
1974
1975         return (error);
1976 }
1977
1978
1979 int mlx4_en_init_netdev(struct mlx4_en_dev *mdev, int port,
1980                         struct mlx4_en_port_profile *prof)
1981 {
1982         struct net_device *dev;
1983         struct mlx4_en_priv *priv;
1984         uint8_t dev_addr[ETHER_ADDR_LEN];
1985         int err;
1986         int i;
1987
1988         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1989         dev = priv->dev = if_alloc(IFT_ETHER);
1990         if (dev == NULL) {
1991                 en_err(priv, "Net device allocation failed\n");
1992                 kfree(priv);
1993                 return -ENOMEM;
1994         }
1995         dev->if_softc = priv;
1996         if_initname(dev, "mlxen", atomic_fetchadd_int(&mlx4_en_unit, 1));
1997         dev->if_mtu = ETHERMTU;
1998         dev->if_init = mlx4_en_open;
1999         dev->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2000         dev->if_ioctl = mlx4_en_ioctl;
2001         dev->if_transmit = mlx4_en_transmit;
2002         dev->if_qflush = mlx4_en_qflush;
2003         dev->if_snd.ifq_maxlen = prof->tx_ring_size;
2004
2005         /*
2006          * Initialize driver private data
2007          */
2008         priv->counter_index = 0xff;
2009         spin_lock_init(&priv->stats_lock);
2010         INIT_WORK(&priv->rx_mode_task, mlx4_en_do_set_rx_mode);
2011         INIT_WORK(&priv->watchdog_task, mlx4_en_restart);
2012         INIT_WORK(&priv->linkstate_task, mlx4_en_linkstate);
2013         INIT_DELAYED_WORK(&priv->stats_task, mlx4_en_do_get_stats);
2014         INIT_DELAYED_WORK(&priv->service_task, mlx4_en_service_task);
2015         callout_init(&priv->watchdog_timer, 1);
2016 #ifdef CONFIG_RFS_ACCEL
2017         INIT_LIST_HEAD(&priv->filters);
2018         spin_lock_init(&priv->filters_lock);
2019 #endif
2020
2021         priv->msg_enable = MLX4_EN_MSG_LEVEL;
2022         priv->dev = dev;
2023         priv->mdev = mdev;
2024         priv->ddev = &mdev->pdev->dev;
2025         priv->prof = prof;
2026         priv->port = port;
2027         priv->port_up = false;
2028         priv->flags = prof->flags;
2029         priv->ctrl_flags = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE |
2030                         MLX4_WQE_CTRL_SOLICITED);
2031
2032         priv->num_tx_rings_p_up = mdev->profile.num_tx_rings_p_up;
2033         priv->tx_ring_num = prof->tx_ring_num;
2034         priv->tx_ring = kcalloc(MAX_TX_RINGS,
2035                                 sizeof(struct mlx4_en_tx_ring *), GFP_KERNEL);
2036         if (!priv->tx_ring) {
2037                 err = -ENOMEM;
2038                 goto out;
2039         }
2040         priv->tx_cq = kcalloc(sizeof(struct mlx4_en_cq *), MAX_TX_RINGS,
2041                         GFP_KERNEL);
2042         if (!priv->tx_cq) {
2043                 err = -ENOMEM;
2044                 goto out;
2045         }
2046
2047         priv->rx_ring_num = prof->rx_ring_num;
2048         priv->cqe_factor = (mdev->dev->caps.cqe_size == 64) ? 1 : 0;
2049         priv->mac_index = -1;
2050         priv->last_ifq_jiffies = 0;
2051         priv->if_counters_rx_errors = 0;
2052         priv->if_counters_rx_no_buffer = 0;
2053 #ifdef CONFIG_MLX4_EN_DCB
2054         if (!mlx4_is_slave(priv->mdev->dev)) {
2055                 priv->dcbx_cap = DCB_CAP_DCBX_HOST;
2056                 priv->flags |= MLX4_EN_FLAG_DCB_ENABLED;
2057                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) {
2058                         dev->dcbnl_ops = &mlx4_en_dcbnl_ops;
2059                 } else {
2060                         en_info(priv, "QoS disabled - no HW support\n");
2061                         dev->dcbnl_ops = &mlx4_en_dcbnl_pfc_ops;
2062                 }
2063         }
2064 #endif
2065
2066         for (i = 0; i < MLX4_EN_MAC_HASH_SIZE; ++i)
2067                 INIT_HLIST_HEAD(&priv->mac_hash[i]);
2068
2069
2070         /* Query for default mac and max mtu */
2071         priv->max_mtu = mdev->dev->caps.eth_mtu_cap[priv->port];
2072         priv->mac = mdev->dev->caps.def_mac[priv->port];
2073         if (ILLEGAL_MAC(priv->mac)) {
2074 #if BITS_PER_LONG == 64
2075                 en_err(priv, "Port: %d, invalid mac burned: 0x%lx, quiting\n",
2076                                 priv->port, priv->mac);
2077 #elif BITS_PER_LONG == 32
2078                 en_err(priv, "Port: %d, invalid mac burned: 0x%llx, quiting\n",
2079                                 priv->port, priv->mac);
2080 #endif
2081                 err = -EINVAL;
2082                 goto out;
2083         }
2084
2085
2086
2087         priv->stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
2088                                           DS_SIZE);
2089
2090         mlx4_en_sysctl_conf(priv);
2091
2092         err = mlx4_en_alloc_resources(priv);
2093         if (err)
2094                 goto out;
2095
2096         /* Allocate page for receive rings */
2097         err = mlx4_alloc_hwq_res(mdev->dev, &priv->res,
2098                                 MLX4_EN_PAGE_SIZE, MLX4_EN_PAGE_SIZE);
2099         if (err) {
2100                 en_err(priv, "Failed to allocate page for rx qps\n");
2101                 goto out;
2102         }
2103         priv->allocated = 1;
2104
2105         /*
2106          * Set driver features
2107          */
2108         dev->if_capabilities |= IFCAP_RXCSUM | IFCAP_TXCSUM;
2109         dev->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
2110         dev->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER;
2111         dev->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU;
2112         dev->if_capabilities |= IFCAP_LRO;
2113
2114         if (mdev->LSO_support)
2115                 dev->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_VLAN_HWTSO;
2116
2117         /* set TSO limits so that we don't have to drop TX packets */
2118         dev->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2119         dev->if_hw_tsomaxsegcount = 16;
2120         dev->if_hw_tsomaxsegsize = 65536;       /* XXX can do up to 4GByte */
2121
2122         dev->if_capenable = dev->if_capabilities;
2123
2124         dev->if_hwassist = 0;
2125         if (dev->if_capenable & (IFCAP_TSO4 | IFCAP_TSO6))
2126                 dev->if_hwassist |= CSUM_TSO;
2127         if (dev->if_capenable & IFCAP_TXCSUM)
2128                 dev->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP);
2129
2130
2131         /* Register for VLAN events */
2132         priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
2133             mlx4_en_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST);
2134         priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
2135             mlx4_en_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST);
2136
2137         mdev->pndev[priv->port] = dev;
2138
2139         priv->last_link_state = MLX4_DEV_EVENT_PORT_DOWN;
2140         mlx4_en_set_default_moderation(priv);
2141
2142         /* Set default MAC */
2143         for (i = 0; i < ETHER_ADDR_LEN; i++)
2144                 dev_addr[ETHER_ADDR_LEN - 1 - i] = (u8) (priv->mac >> (8 * i));
2145
2146
2147         ether_ifattach(dev, dev_addr);
2148         if_link_state_change(dev, LINK_STATE_DOWN);
2149         ifmedia_init(&priv->media, IFM_IMASK | IFM_ETH_FMASK,
2150             mlx4_en_media_change, mlx4_en_media_status);
2151         ifmedia_add(&priv->media, IFM_ETHER | IFM_FDX | IFM_1000_T, 0, NULL);
2152         ifmedia_add(&priv->media, IFM_ETHER | IFM_FDX | IFM_10G_SR, 0, NULL);
2153         ifmedia_add(&priv->media, IFM_ETHER | IFM_FDX | IFM_10G_CX4, 0, NULL);
2154         ifmedia_add(&priv->media, IFM_ETHER | IFM_FDX | IFM_40G_CR4, 0, NULL);
2155         ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2156         ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO);
2157
2158         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2159         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2160
2161         priv->registered = 1;
2162
2163         en_warn(priv, "Using %d TX rings\n", prof->tx_ring_num);
2164         en_warn(priv, "Using %d RX rings\n", prof->rx_ring_num);
2165
2166
2167         priv->rx_mb_size = dev->if_mtu + ETH_HLEN + VLAN_HLEN + ETH_FCS_LEN;
2168         err = mlx4_SET_PORT_general(mdev->dev, priv->port,
2169                                     priv->rx_mb_size,
2170                                     prof->tx_pause, prof->tx_ppp,
2171                                     prof->rx_pause, prof->rx_ppp);
2172         if (err) {
2173                 en_err(priv, "Failed setting port general configurations "
2174                        "for port %d, with error %d\n", priv->port, err);
2175                 goto out;
2176         }
2177
2178         /* Init port */
2179         en_warn(priv, "Initializing port\n");
2180         err = mlx4_INIT_PORT(mdev->dev, priv->port);
2181         if (err) {
2182                 en_err(priv, "Failed Initializing port\n");
2183                 goto out;
2184         }
2185
2186         queue_delayed_work(mdev->workqueue, &priv->stats_task, STATS_DELAY);
2187
2188         if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS)
2189                 queue_delayed_work(mdev->workqueue, &priv->service_task, SERVICE_TASK_DELAY);
2190
2191
2192
2193         return 0;
2194
2195 out:
2196         mlx4_en_destroy_netdev(dev);
2197         return err;
2198 }
2199
2200 static int mlx4_en_set_ring_size(struct net_device *dev,
2201     int rx_size, int tx_size)
2202 {
2203         struct mlx4_en_priv *priv = netdev_priv(dev);
2204         struct mlx4_en_dev *mdev = priv->mdev;
2205         int port_up = 0;
2206         int err = 0;
2207
2208         rx_size = roundup_pow_of_two(rx_size);
2209         rx_size = max_t(u32, rx_size, MLX4_EN_MIN_RX_SIZE);
2210         rx_size = min_t(u32, rx_size, MLX4_EN_MAX_RX_SIZE);
2211         tx_size = roundup_pow_of_two(tx_size);
2212         tx_size = max_t(u32, tx_size, MLX4_EN_MIN_TX_SIZE);
2213         tx_size = min_t(u32, tx_size, MLX4_EN_MAX_TX_SIZE);
2214
2215         if (rx_size == (priv->port_up ?
2216             priv->rx_ring[0]->actual_size : priv->rx_ring[0]->size) &&
2217             tx_size == priv->tx_ring[0]->size)
2218                 return 0;
2219         mutex_lock(&mdev->state_lock);
2220         if (priv->port_up) {
2221                 port_up = 1;
2222                 mlx4_en_stop_port(dev);
2223         }
2224         mlx4_en_free_resources(priv);
2225         priv->prof->tx_ring_size = tx_size;
2226         priv->prof->rx_ring_size = rx_size;
2227         err = mlx4_en_alloc_resources(priv);
2228         if (err) {
2229                 en_err(priv, "Failed reallocating port resources\n");
2230                 goto out;
2231         }
2232         if (port_up) {
2233                 err = mlx4_en_start_port(dev);
2234                 if (err)
2235                         en_err(priv, "Failed starting port\n");
2236         }
2237 out:
2238         mutex_unlock(&mdev->state_lock);
2239         return err;
2240 }
2241 static int mlx4_en_set_rx_ring_size(SYSCTL_HANDLER_ARGS)
2242 {
2243         struct mlx4_en_priv *priv;
2244         int size;
2245         int error;
2246
2247         priv = arg1;
2248         size = priv->prof->rx_ring_size;
2249         error = sysctl_handle_int(oidp, &size, 0, req);
2250         if (error || !req->newptr)
2251                 return (error);
2252         error = -mlx4_en_set_ring_size(priv->dev, size,
2253             priv->prof->tx_ring_size);
2254         return (error);
2255 }
2256
2257 static int mlx4_en_set_tx_ring_size(SYSCTL_HANDLER_ARGS)
2258 {
2259         struct mlx4_en_priv *priv;
2260         int size;
2261         int error;
2262
2263         priv = arg1;
2264         size = priv->prof->tx_ring_size;
2265         error = sysctl_handle_int(oidp, &size, 0, req);
2266         if (error || !req->newptr)
2267                 return (error);
2268         error = -mlx4_en_set_ring_size(priv->dev, priv->prof->rx_ring_size,
2269             size);
2270
2271         return (error);
2272 }
2273
2274 static int mlx4_en_set_tx_ppp(SYSCTL_HANDLER_ARGS)
2275 {
2276         struct mlx4_en_priv *priv;
2277         int ppp;
2278         int error;
2279
2280         priv = arg1;
2281         ppp = priv->prof->tx_ppp;
2282         error = sysctl_handle_int(oidp, &ppp, 0, req);
2283         if (error || !req->newptr)
2284                 return (error);
2285         if (ppp > 0xff || ppp < 0)
2286                 return (-EINVAL);
2287         priv->prof->tx_ppp = ppp;
2288         error = -mlx4_SET_PORT_general(priv->mdev->dev, priv->port,
2289                                        priv->rx_mb_size + ETHER_CRC_LEN,
2290                                        priv->prof->tx_pause,
2291                                        priv->prof->tx_ppp,
2292                                        priv->prof->rx_pause,
2293                                        priv->prof->rx_ppp);
2294
2295         return (error);
2296 }
2297
2298 static int mlx4_en_set_rx_ppp(SYSCTL_HANDLER_ARGS)
2299 {
2300         struct mlx4_en_priv *priv;
2301         struct mlx4_en_dev *mdev;
2302         int ppp;
2303         int error;
2304         int port_up;
2305
2306         port_up = 0;
2307         priv = arg1;
2308         mdev = priv->mdev;
2309         ppp = priv->prof->rx_ppp;
2310         error = sysctl_handle_int(oidp, &ppp, 0, req);
2311         if (error || !req->newptr)
2312                 return (error);
2313         if (ppp > 0xff || ppp < 0)
2314                 return (-EINVAL);
2315         /* See if we have to change the number of tx queues. */
2316         if (!ppp != !priv->prof->rx_ppp) {
2317                 mutex_lock(&mdev->state_lock);
2318                 if (priv->port_up) {
2319                         port_up = 1;
2320                         mlx4_en_stop_port(priv->dev);
2321                 }
2322                 mlx4_en_free_resources(priv);
2323                 priv->prof->rx_ppp = ppp;
2324                 error = -mlx4_en_alloc_resources(priv);
2325                 if (error)
2326                         en_err(priv, "Failed reallocating port resources\n");
2327                 if (error == 0 && port_up) {
2328                         error = -mlx4_en_start_port(priv->dev);
2329                         if (error)
2330                                 en_err(priv, "Failed starting port\n");
2331                 }
2332                 mutex_unlock(&mdev->state_lock);
2333                 return (error);
2334
2335         }
2336         priv->prof->rx_ppp = ppp;
2337         error = -mlx4_SET_PORT_general(priv->mdev->dev, priv->port,
2338                                        priv->rx_mb_size + ETHER_CRC_LEN,
2339                                        priv->prof->tx_pause,
2340                                        priv->prof->tx_ppp,
2341                                        priv->prof->rx_pause,
2342                                        priv->prof->rx_ppp);
2343
2344         return (error);
2345 }
2346
2347 static void mlx4_en_sysctl_conf(struct mlx4_en_priv *priv)
2348 {
2349         struct net_device *dev;
2350         struct sysctl_ctx_list *ctx;
2351         struct sysctl_oid *node;
2352         struct sysctl_oid_list *node_list;
2353         struct sysctl_oid *coal;
2354         struct sysctl_oid_list *coal_list;
2355         const char *pnameunit;
2356
2357         dev = priv->dev;
2358         ctx = &priv->conf_ctx;
2359         pnameunit = device_get_nameunit(priv->mdev->pdev->dev.bsddev);
2360
2361         sysctl_ctx_init(ctx);
2362         priv->sysctl = SYSCTL_ADD_NODE(ctx, SYSCTL_STATIC_CHILDREN(_hw),
2363             OID_AUTO, dev->if_xname, CTLFLAG_RD, 0, "mlx4 10gig ethernet");
2364         node = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(priv->sysctl), OID_AUTO,
2365             "conf", CTLFLAG_RD, NULL, "Configuration");
2366         node_list = SYSCTL_CHILDREN(node);
2367
2368         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "msg_enable",
2369             CTLFLAG_RW, &priv->msg_enable, 0,
2370             "Driver message enable bitfield");
2371         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "rx_rings",
2372             CTLFLAG_RD, &priv->rx_ring_num, 0,
2373             "Number of receive rings");
2374         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "tx_rings",
2375             CTLFLAG_RD, &priv->tx_ring_num, 0,
2376             "Number of transmit rings");
2377         SYSCTL_ADD_PROC(ctx, node_list, OID_AUTO, "rx_size",
2378             CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, priv, 0,
2379             mlx4_en_set_rx_ring_size, "I", "Receive ring size");
2380         SYSCTL_ADD_PROC(ctx, node_list, OID_AUTO, "tx_size",
2381             CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, priv, 0,
2382             mlx4_en_set_tx_ring_size, "I", "Transmit ring size");
2383         SYSCTL_ADD_PROC(ctx, node_list, OID_AUTO, "tx_ppp",
2384             CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, priv, 0,
2385             mlx4_en_set_tx_ppp, "I", "TX Per-priority pause");
2386         SYSCTL_ADD_PROC(ctx, node_list, OID_AUTO, "rx_ppp",
2387             CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, priv, 0,
2388             mlx4_en_set_rx_ppp, "I", "RX Per-priority pause");
2389         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "port_num",
2390             CTLFLAG_RD, &priv->port, 0,
2391             "Port Number");
2392         SYSCTL_ADD_STRING(ctx, node_list, OID_AUTO, "device_name",
2393             CTLFLAG_RD, __DECONST(void *, pnameunit), 0,
2394             "PCI device name");
2395
2396         /* Add coalescer configuration. */
2397         coal = SYSCTL_ADD_NODE(ctx, node_list, OID_AUTO,
2398             "coalesce", CTLFLAG_RD, NULL, "Interrupt coalesce configuration");
2399         coal_list = SYSCTL_CHILDREN(node);
2400         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "pkt_rate_low",
2401             CTLFLAG_RW, &priv->pkt_rate_low, 0,
2402             "Packets per-second for minimum delay");
2403         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "rx_usecs_low",
2404             CTLFLAG_RW, &priv->rx_usecs_low, 0,
2405             "Minimum RX delay in micro-seconds");
2406         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "pkt_rate_high",
2407             CTLFLAG_RW, &priv->pkt_rate_high, 0,
2408             "Packets per-second for maximum delay");
2409         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "rx_usecs_high",
2410             CTLFLAG_RW, &priv->rx_usecs_high, 0,
2411             "Maximum RX delay in micro-seconds");
2412         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "sample_interval",
2413             CTLFLAG_RW, &priv->sample_interval, 0,
2414             "adaptive frequency in units of HZ ticks");
2415         SYSCTL_ADD_UINT(ctx, coal_list, OID_AUTO, "adaptive_rx_coal",
2416             CTLFLAG_RW, &priv->adaptive_rx_coal, 0,
2417             "Enable adaptive rx coalescing");
2418 }
2419
2420 static void mlx4_en_sysctl_stat(struct mlx4_en_priv *priv)
2421 {
2422         struct net_device *dev;
2423         struct sysctl_ctx_list *ctx;
2424         struct sysctl_oid *node;
2425         struct sysctl_oid_list *node_list;
2426         struct sysctl_oid *ring_node;
2427         struct sysctl_oid_list *ring_list;
2428         struct mlx4_en_tx_ring *tx_ring;
2429         struct mlx4_en_rx_ring *rx_ring;
2430         char namebuf[128];
2431         int i;
2432
2433         dev = priv->dev;
2434
2435         ctx = &priv->stat_ctx;
2436         sysctl_ctx_init(ctx);
2437         node = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(priv->sysctl), OID_AUTO,
2438             "stat", CTLFLAG_RD, NULL, "Statistics");
2439         node_list = SYSCTL_CHILDREN(node);
2440
2441 #ifdef MLX4_EN_PERF_STAT
2442         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "tx_poll", CTLFLAG_RD,
2443             &priv->pstats.tx_poll, "TX Poll calls");
2444         SYSCTL_ADD_QUAD(ctx, node_list, OID_AUTO, "tx_pktsz_avg", CTLFLAG_RD,
2445             &priv->pstats.tx_pktsz_avg, "TX average packet size");
2446         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "inflight_avg", CTLFLAG_RD,
2447             &priv->pstats.inflight_avg, "TX average packets in-flight");
2448         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "tx_coal_avg", CTLFLAG_RD,
2449             &priv->pstats.tx_coal_avg, "TX average coalesced completions");
2450         SYSCTL_ADD_UINT(ctx, node_list, OID_AUTO, "rx_coal_avg", CTLFLAG_RD,
2451             &priv->pstats.rx_coal_avg, "RX average coalesced completions");
2452 #endif
2453
2454         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tso_packets", CTLFLAG_RD,
2455             &priv->port_stats.tso_packets, "TSO packets sent");
2456         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "queue_stopped", CTLFLAG_RD,
2457             &priv->port_stats.queue_stopped, "Queue full");
2458         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "wake_queue", CTLFLAG_RD,
2459             &priv->port_stats.wake_queue, "Queue resumed after full");
2460         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_timeout", CTLFLAG_RD,
2461             &priv->port_stats.tx_timeout, "Transmit timeouts");
2462         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_alloc_failed", CTLFLAG_RD,
2463             &priv->port_stats.rx_alloc_failed, "RX failed to allocate mbuf");
2464         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_chksum_good", CTLFLAG_RD,
2465             &priv->port_stats.rx_chksum_good, "RX checksum offload success");
2466         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_chksum_none", CTLFLAG_RD,
2467             &priv->port_stats.rx_chksum_none, "RX without checksum offload");
2468         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_chksum_offload",
2469             CTLFLAG_RD, &priv->port_stats.tx_chksum_offload,
2470             "TX checksum offloads");
2471
2472         /* Could strdup the names and add in a loop.  This is simpler. */
2473         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_bytes", CTLFLAG_RD,
2474             &priv->pkstats.rx_bytes, "RX Bytes");
2475         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_packets", CTLFLAG_RD,
2476             &priv->pkstats.rx_packets, "RX packets");
2477         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_multicast_packets", CTLFLAG_RD,
2478             &priv->pkstats.rx_multicast_packets, "RX Multicast Packets");
2479         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_broadcast_packets", CTLFLAG_RD,
2480             &priv->pkstats.rx_broadcast_packets, "RX Broadcast Packets");
2481         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_errors", CTLFLAG_RD,
2482             &priv->pkstats.rx_errors, "RX Errors");
2483         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_dropped", CTLFLAG_RD,
2484             &priv->pkstats.rx_dropped, "RX Dropped");
2485         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_length_errors", CTLFLAG_RD,
2486             &priv->pkstats.rx_length_errors, "RX Length Errors");
2487         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_over_errors", CTLFLAG_RD,
2488             &priv->pkstats.rx_over_errors, "RX Over Errors");
2489         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_crc_errors", CTLFLAG_RD,
2490             &priv->pkstats.rx_crc_errors, "RX CRC Errors");
2491         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_jabbers", CTLFLAG_RD,
2492             &priv->pkstats.rx_jabbers, "RX Jabbers");
2493
2494
2495         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_in_range_length_error", CTLFLAG_RD,
2496             &priv->pkstats.rx_in_range_length_error, "RX IN_Range Length Error");
2497         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_out_range_length_error",
2498                 CTLFLAG_RD, &priv->pkstats.rx_out_range_length_error,
2499                 "RX Out Range Length Error");
2500         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_lt_64_bytes_packets", CTLFLAG_RD,
2501             &priv->pkstats.rx_lt_64_bytes_packets, "RX Lt 64 Bytes Packets");
2502         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_127_bytes_packets", CTLFLAG_RD,
2503             &priv->pkstats.rx_127_bytes_packets, "RX 127 bytes Packets");
2504         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_255_bytes_packets", CTLFLAG_RD,
2505             &priv->pkstats.rx_255_bytes_packets, "RX 255 bytes Packets");
2506         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_511_bytes_packets", CTLFLAG_RD,
2507             &priv->pkstats.rx_511_bytes_packets, "RX 511 bytes Packets");
2508         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_1023_bytes_packets", CTLFLAG_RD,
2509             &priv->pkstats.rx_1023_bytes_packets, "RX 1023 bytes Packets");
2510         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_1518_bytes_packets", CTLFLAG_RD,
2511             &priv->pkstats.rx_1518_bytes_packets, "RX 1518 bytes Packets");
2512         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_1522_bytes_packets", CTLFLAG_RD,
2513             &priv->pkstats.rx_1522_bytes_packets, "RX 1522 bytes Packets");
2514         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_1548_bytes_packets", CTLFLAG_RD,
2515             &priv->pkstats.rx_1548_bytes_packets, "RX 1548 bytes Packets");
2516         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "rx_gt_1548_bytes_packets", CTLFLAG_RD,
2517             &priv->pkstats.rx_gt_1548_bytes_packets,
2518             "RX Greater Then 1548 bytes Packets");
2519
2520 struct mlx4_en_pkt_stats {
2521         unsigned long tx_packets;
2522         unsigned long tx_bytes;
2523         unsigned long tx_multicast_packets;
2524         unsigned long tx_broadcast_packets;
2525         unsigned long tx_errors;
2526         unsigned long tx_dropped;
2527         unsigned long tx_lt_64_bytes_packets;
2528         unsigned long tx_127_bytes_packets;
2529         unsigned long tx_255_bytes_packets;
2530         unsigned long tx_511_bytes_packets;
2531         unsigned long tx_1023_bytes_packets;
2532         unsigned long tx_1518_bytes_packets;
2533         unsigned long tx_1522_bytes_packets;
2534         unsigned long tx_1548_bytes_packets;
2535         unsigned long tx_gt_1548_bytes_packets;
2536         unsigned long rx_prio[NUM_PRIORITIES][NUM_PRIORITY_STATS];
2537         unsigned long tx_prio[NUM_PRIORITIES][NUM_PRIORITY_STATS];
2538 #define NUM_PKT_STATS           72
2539 };
2540
2541
2542         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_packets", CTLFLAG_RD,
2543             &priv->pkstats.tx_packets, "TX packets");
2544         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_bytes", CTLFLAG_RD,
2545             &priv->pkstats.tx_packets, "TX Bytes");
2546         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_multicast_packets", CTLFLAG_RD,
2547             &priv->pkstats.tx_multicast_packets, "TX Multicast Packets");
2548         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_broadcast_packets", CTLFLAG_RD,
2549             &priv->pkstats.tx_broadcast_packets, "TX Broadcast Packets");
2550         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_errors", CTLFLAG_RD,
2551             &priv->pkstats.tx_errors, "TX Errors");
2552         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_dropped", CTLFLAG_RD,
2553             &priv->pkstats.tx_dropped, "TX Dropped");
2554         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_lt_64_bytes_packets", CTLFLAG_RD,
2555             &priv->pkstats.tx_lt_64_bytes_packets, "TX Less Then 64 Bytes Packets");
2556         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_127_bytes_packets", CTLFLAG_RD,
2557             &priv->pkstats.tx_127_bytes_packets, "TX 127 Bytes Packets");
2558         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_255_bytes_packets", CTLFLAG_RD,
2559             &priv->pkstats.tx_255_bytes_packets, "TX 255 Bytes Packets");
2560         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_511_bytes_packets", CTLFLAG_RD,
2561             &priv->pkstats.tx_511_bytes_packets, "TX 511 Bytes Packets");
2562         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_1023_bytes_packets", CTLFLAG_RD,
2563             &priv->pkstats.tx_1023_bytes_packets, "TX 1023 Bytes Packets");
2564         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_1518_bytes_packets", CTLFLAG_RD,
2565             &priv->pkstats.tx_1518_bytes_packets, "TX 1518 Bytes Packets");
2566         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_1522_bytes_packets", CTLFLAG_RD,
2567             &priv->pkstats.tx_1522_bytes_packets, "TX 1522 Bytes Packets");
2568         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_1548_bytes_packets", CTLFLAG_RD,
2569             &priv->pkstats.tx_1548_bytes_packets, "TX 1548 Bytes Packets");
2570         SYSCTL_ADD_ULONG(ctx, node_list, OID_AUTO, "tx_gt_1548_bytes_packets", CTLFLAG_RD,
2571             &priv->pkstats.tx_gt_1548_bytes_packets,
2572             "TX Greater Then 1548 Bytes Packets");
2573
2574
2575
2576         for (i = 0; i < priv->tx_ring_num; i++) {
2577                 tx_ring = priv->tx_ring[i];
2578                 snprintf(namebuf, sizeof(namebuf), "tx_ring%d", i);
2579                 ring_node = SYSCTL_ADD_NODE(ctx, node_list, OID_AUTO, namebuf,
2580                     CTLFLAG_RD, NULL, "TX Ring");
2581                 ring_list = SYSCTL_CHILDREN(ring_node);
2582                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "packets",
2583                     CTLFLAG_RD, &tx_ring->packets, "TX packets");
2584                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "bytes",
2585                     CTLFLAG_RD, &tx_ring->bytes, "TX bytes");
2586
2587         }
2588         for (i = 0; i < priv->rx_ring_num; i++) {
2589                 rx_ring = priv->rx_ring[i];
2590                 snprintf(namebuf, sizeof(namebuf), "rx_ring%d", i);
2591                 ring_node = SYSCTL_ADD_NODE(ctx, node_list, OID_AUTO, namebuf,
2592                     CTLFLAG_RD, NULL, "RX Ring");
2593                 ring_list = SYSCTL_CHILDREN(ring_node);
2594                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "packets",
2595                     CTLFLAG_RD, &rx_ring->packets, "RX packets");
2596                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "bytes",
2597                     CTLFLAG_RD, &rx_ring->bytes, "RX bytes");
2598                 SYSCTL_ADD_ULONG(ctx, ring_list, OID_AUTO, "error",
2599                     CTLFLAG_RD, &rx_ring->errors, "RX soft errors");
2600         }
2601 }