drivers/infiniband/ulp/ipoib/ipoib_ib.c

   1 /*
   2  * Copyright (c) 2004, 2005 Topspin Communications.  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  * $Id: ipoib_ib.c 1386 2004-12-27 16:23:17Z roland $
  33  */
  34
  35 #include <linux/delay.h>
  36 #include <linux/dma-mapping.h>
  37
  38 #include <ib_cache.h>
  39
  40 #include "ipoib.h"
  41
  42 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
  43 static int data_debug_level;
  44
  45 module_param(data_debug_level, int, 0644);
  46 MODULE_PARM_DESC(data_debug_level,
  47                  "Enable data path debug tracing if > 0");
  48 #endif
  49
  50 #define IPOIB_OP_RECV   (1ul << 31)
  51
  52 static DECLARE_MUTEX(pkey_sem);
  53
  54 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
  55                                  struct ib_pd *pd, struct ib_ah_attr *attr)
  56 {
  57         struct ipoib_ah *ah;
  58
  59         ah = kmalloc(sizeof *ah, GFP_KERNEL);
  60         if (!ah)
  61                 return NULL;
  62
  63         ah->dev       = dev;
  64         ah->last_send = 0;
  65         kref_init(&ah->ref);
  66
  67         ah->ah = ib_create_ah(pd, attr);
  68         if (IS_ERR(ah->ah)) {
  69                 kfree(ah);
  70                 ah = NULL;
  71         } else
  72                 ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah);
  73
  74         return ah;
  75 }
  76
  77 void ipoib_free_ah(struct kref *kref)
  78 {
  79         struct ipoib_ah *ah = container_of(kref, struct ipoib_ah, ref);
  80         struct ipoib_dev_priv *priv = netdev_priv(ah->dev);
  81
  82         unsigned long flags;
  83
  84         if (ah->last_send <= priv->tx_tail) {
  85                 ipoib_dbg(priv, "Freeing ah %p\n", ah->ah);
  86                 ib_destroy_ah(ah->ah);
  87                 kfree(ah);
  88         } else {
  89                 spin_lock_irqsave(&priv->lock, flags);
  90                 list_add_tail(&ah->list, &priv->dead_ahs);
  91                 spin_unlock_irqrestore(&priv->lock, flags);
  92         }
  93 }
  94
  95 static inline int ipoib_ib_receive(struct ipoib_dev_priv *priv,
  96                                    unsigned int wr_id,
  97                                    dma_addr_t addr)
  98 {
  99         struct ib_sge list = {
 100                 .addr    = addr,
 101                 .length  = IPOIB_BUF_SIZE,
 102                 .lkey    = priv->mr->lkey,
 103         };
 104         struct ib_recv_wr param = {
 105                 .wr_id      = wr_id | IPOIB_OP_RECV,
 106                 .sg_list    = &list,
 107                 .num_sge    = 1,
 108         };
 109         struct ib_recv_wr *bad_wr;
 110
 111         return ib_post_recv(priv->qp, &param, &bad_wr);
 112 }
 113
 114 static int ipoib_ib_post_receive(struct net_device *dev, int id)
 115 {
 116         struct ipoib_dev_priv *priv = netdev_priv(dev);
 117         struct sk_buff *skb;
 118         dma_addr_t addr;
 119         int ret;
 120
 121         skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4);
 122         if (!skb) {
 123                 ipoib_warn(priv, "failed to allocate receive buffer\n");
 124
 125                 priv->rx_ring[id].skb = NULL;
 126                 return -ENOMEM;
 127         }
 128         skb_reserve(skb, 4);    /* 16 byte align IP header */
 129         priv->rx_ring[id].skb = skb;
 130         addr = dma_map_single(priv->ca->dma_device,
 131                               skb->data, IPOIB_BUF_SIZE,
 132                               DMA_FROM_DEVICE);
 133         pci_unmap_addr_set(&priv->rx_ring[id], mapping, addr);
 134
 135         ret = ipoib_ib_receive(priv, id, addr);
 136         if (ret) {
 137                 ipoib_warn(priv, "ipoib_ib_receive failed for buf %d (%d)\n",
 138                            id, ret);
 139                 dma_unmap_single(priv->ca->dma_device, addr,
 140                                  IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
 141                 dev_kfree_skb_any(skb);
 142                 priv->rx_ring[id].skb = NULL;
 143         }
 144
 145         return ret;
 146 }
 147
 148 static int ipoib_ib_post_receives(struct net_device *dev)
 149 {
 150         struct ipoib_dev_priv *priv = netdev_priv(dev);
 151         int i;
 152
 153         for (i = 0; i < IPOIB_RX_RING_SIZE; ++i) {
 154                 if (ipoib_ib_post_receive(dev, i)) {
 155                         ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i);
 156                         return -EIO;
 157                 }
 158         }
 159
 160         return 0;
 161 }
 162
 163 static void ipoib_ib_handle_wc(struct net_device *dev,
 164                                struct ib_wc *wc)
 165 {
 166         struct ipoib_dev_priv *priv = netdev_priv(dev);
 167         unsigned int wr_id = wc->wr_id;
 168
 169         ipoib_dbg_data(priv, "called: id %d, op %d, status: %d\n",
 170                        wr_id, wc->opcode, wc->status);
 171
 172         if (wr_id & IPOIB_OP_RECV) {
 173                 wr_id &= ~IPOIB_OP_RECV;
 174
 175                 if (wr_id < IPOIB_RX_RING_SIZE) {
 176                         struct sk_buff *skb = priv->rx_ring[wr_id].skb;
 177
 178                         priv->rx_ring[wr_id].skb = NULL;
 179
 180                         dma_unmap_single(priv->ca->dma_device,
 181                                          pci_unmap_addr(&priv->rx_ring[wr_id],
 182                                                         mapping),
 183                                          IPOIB_BUF_SIZE,
 184                                          DMA_FROM_DEVICE);
 185
 186                         if (wc->status != IB_WC_SUCCESS) {
 187                                 if (wc->status != IB_WC_WR_FLUSH_ERR)
 188                                         ipoib_warn(priv, "failed recv event "
 189                                                    "(status=%d, wrid=%d vend_err %x)\n",
 190                                                    wc->status, wr_id, wc->vendor_err);
 191                                 dev_kfree_skb_any(skb);
 192                                 return;
 193                         }
 194
 195                         ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
 196                                        wc->byte_len, wc->slid);
 197
 198                         skb_put(skb, wc->byte_len);
 199                         skb_pull(skb, IB_GRH_BYTES);
 200
 201                         if (wc->slid != priv->local_lid ||
 202                             wc->src_qp != priv->qp->qp_num) {
 203                                 skb->protocol = ((struct ipoib_header *) skb->data)->proto;
 204
 205                                 skb_pull(skb, IPOIB_ENCAP_LEN);
 206
 207                                 dev->last_rx = jiffies;
 208                                 ++priv->stats.rx_packets;
 209                                 priv->stats.rx_bytes += skb->len;
 210
 211                                 skb->dev = dev;
 212                                 /* XXX get correct PACKET_ type here */
 213                                 skb->pkt_type = PACKET_HOST;
 214                                 netif_rx_ni(skb);
 215                         } else {
 216                                 ipoib_dbg_data(priv, "dropping loopback packet\n");
 217                                 dev_kfree_skb_any(skb);
 218                         }
 219
 220                         /* repost receive */
 221                         if (ipoib_ib_post_receive(dev, wr_id))
 222                                 ipoib_warn(priv, "ipoib_ib_post_receive failed "
 223                                            "for buf %d\n", wr_id);
 224                 } else
 225                         ipoib_warn(priv, "completion event with wrid %d\n",
 226                                    wr_id);
 227
 228         } else {
 229                 struct ipoib_buf *tx_req;
 230                 unsigned long flags;
 231
 232                 if (wr_id >= IPOIB_TX_RING_SIZE) {
 233                         ipoib_warn(priv, "completion event with wrid %d (> %d)\n",
 234                                    wr_id, IPOIB_TX_RING_SIZE);
 235                         return;
 236                 }
 237
 238                 ipoib_dbg_data(priv, "send complete, wrid %d\n", wr_id);
 239
 240                 tx_req = &priv->tx_ring[wr_id];
 241
 242                 dma_unmap_single(priv->ca->dma_device,
 243                                  pci_unmap_addr(tx_req, mapping),
 244                                  tx_req->skb->len,
 245                                  DMA_TO_DEVICE);
 246
 247                 ++priv->stats.tx_packets;
 248                 priv->stats.tx_bytes += tx_req->skb->len;
 249
 250                 dev_kfree_skb_any(tx_req->skb);
 251
 252                 spin_lock_irqsave(&priv->tx_lock, flags);
 253                 ++priv->tx_tail;
 254                 if (netif_queue_stopped(dev) &&
 255                     priv->tx_head - priv->tx_tail <= IPOIB_TX_RING_SIZE / 2)
 256                         netif_wake_queue(dev);
 257                 spin_unlock_irqrestore(&priv->tx_lock, flags);
 258
 259                 if (wc->status != IB_WC_SUCCESS &&
 260                     wc->status != IB_WC_WR_FLUSH_ERR)
 261                         ipoib_warn(priv, "failed send event "
 262                                    "(status=%d, wrid=%d vend_err %x)\n",
 263                                    wc->status, wr_id, wc->vendor_err);
 264         }
 265 }
 266
 267 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
 268 {
 269         struct net_device *dev = (struct net_device *) dev_ptr;
 270         struct ipoib_dev_priv *priv = netdev_priv(dev);
 271         int n, i;
 272
 273         ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
 274         do {
 275                 n = ib_poll_cq(cq, IPOIB_NUM_WC, priv->ibwc);
 276                 for (i = 0; i < n; ++i)
 277                         ipoib_ib_handle_wc(dev, priv->ibwc + i);
 278         } while (n == IPOIB_NUM_WC);
 279 }
 280
 281 static inline int post_send(struct ipoib_dev_priv *priv,
 282                             unsigned int wr_id,
 283                             struct ib_ah *address, u32 qpn,
 284                             dma_addr_t addr, int len)
 285 {
 286         struct ib_send_wr *bad_wr;
 287
 288         priv->tx_sge.addr             = addr;
 289         priv->tx_sge.length           = len;
 290
 291         priv->tx_wr.wr_id             = wr_id;
 292         priv->tx_wr.wr.ud.remote_qpn  = qpn;
 293         priv->tx_wr.wr.ud.ah          = address;
 294
 295         return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr);
 296 }
 297
 298 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 299                 struct ipoib_ah *address, u32 qpn)
 300 {
 301         struct ipoib_dev_priv *priv = netdev_priv(dev);
 302         struct ipoib_buf *tx_req;
 303         dma_addr_t addr;
 304
 305         if (skb->len > dev->mtu + INFINIBAND_ALEN) {
 306                 ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
 307                            skb->len, dev->mtu + INFINIBAND_ALEN);
 308                 ++priv->stats.tx_dropped;
 309                 ++priv->stats.tx_errors;
 310                 dev_kfree_skb_any(skb);
 311                 return;
 312         }
 313
 314         ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n",
 315                        skb->len, address, qpn);
 316
 317         /*
 318          * We put the skb into the tx_ring _before_ we call post_send()
 319          * because it's entirely possible that the completion handler will
 320          * run before we execute anything after the post_send().  That
 321          * means we have to make sure everything is properly recorded and
 322          * our state is consistent before we call post_send().
 323          */
 324         tx_req = &priv->tx_ring[priv->tx_head & (IPOIB_TX_RING_SIZE - 1)];
 325         tx_req->skb = skb;
 326         addr = dma_map_single(priv->ca->dma_device, skb->data, skb->len,
 327                               DMA_TO_DEVICE);
 328         pci_unmap_addr_set(tx_req, mapping, addr);
 329
 330         if (unlikely(post_send(priv, priv->tx_head & (IPOIB_TX_RING_SIZE - 1),
 331                                address->ah, qpn, addr, skb->len))) {
 332                 ipoib_warn(priv, "post_send failed\n");
 333                 ++priv->stats.tx_errors;
 334                 dma_unmap_single(priv->ca->dma_device, addr, skb->len,
 335                                  DMA_TO_DEVICE);
 336                 dev_kfree_skb_any(skb);
 337         } else {
 338                 dev->trans_start = jiffies;
 339
 340                 address->last_send = priv->tx_head;
 341                 ++priv->tx_head;
 342
 343                 if (priv->tx_head - priv->tx_tail == IPOIB_TX_RING_SIZE) {
 344                         ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
 345                         netif_stop_queue(dev);
 346                 }
 347         }
 348 }
 349
 350 static void __ipoib_reap_ah(struct net_device *dev)
 351 {
 352         struct ipoib_dev_priv *priv = netdev_priv(dev);
 353         struct ipoib_ah *ah, *tah;
 354         LIST_HEAD(remove_list);
 355
 356         spin_lock_irq(&priv->lock);
 357         list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list)
 358                 if (ah->last_send <= priv->tx_tail) {
 359                         list_del(&ah->list);
 360                         list_add_tail(&ah->list, &remove_list);
 361                 }
 362         spin_unlock_irq(&priv->lock);
 363
 364         list_for_each_entry_safe(ah, tah, &remove_list, list) {
 365                 ipoib_dbg(priv, "Reaping ah %p\n", ah->ah);
 366                 ib_destroy_ah(ah->ah);
 367                 kfree(ah);
 368         }
 369 }
 370
 371 void ipoib_reap_ah(void *dev_ptr)
 372 {
 373         struct net_device *dev = dev_ptr;
 374         struct ipoib_dev_priv *priv = netdev_priv(dev);
 375
 376         __ipoib_reap_ah(dev);
 377
 378         if (!test_bit(IPOIB_STOP_REAPER, &priv->flags))
 379                 queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
 380 }
 381
 382 int ipoib_ib_dev_open(struct net_device *dev)
 383 {
 384         struct ipoib_dev_priv *priv = netdev_priv(dev);
 385         int ret;
 386
 387         ret = ipoib_qp_create(dev);
 388         if (ret) {
 389                 ipoib_warn(priv, "ipoib_qp_create returned %d\n", ret);
 390                 return -1;
 391         }
 392
 393         ret = ipoib_ib_post_receives(dev);
 394         if (ret) {
 395                 ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret);
 396                 return -1;
 397         }
 398
 399         clear_bit(IPOIB_STOP_REAPER, &priv->flags);
 400         queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
 401
 402         return 0;
 403 }
 404
 405 int ipoib_ib_dev_up(struct net_device *dev)
 406 {
 407         struct ipoib_dev_priv *priv = netdev_priv(dev);
 408
 409         set_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
 410
 411         return ipoib_mcast_start_thread(dev);
 412 }
 413
 414 int ipoib_ib_dev_down(struct net_device *dev)
 415 {
 416         struct ipoib_dev_priv *priv = netdev_priv(dev);
 417
 418         ipoib_dbg(priv, "downing ib_dev\n");
 419
 420         clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
 421         netif_carrier_off(dev);
 422
 423         /* Shutdown the P_Key thread if still active */
 424         if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
 425                 down(&pkey_sem);
 426                 set_bit(IPOIB_PKEY_STOP, &priv->flags);
 427                 cancel_delayed_work(&priv->pkey_task);
 428                 up(&pkey_sem);
 429                 flush_workqueue(ipoib_workqueue);
 430         }
 431
 432         ipoib_mcast_stop_thread(dev);
 433
 434         /*
 435          * Flush the multicast groups first so we stop any multicast joins. The
 436          * completion thread may have already died and we may deadlock waiting
 437          * for the completion thread to finish some multicast joins.
 438          */
 439         ipoib_mcast_dev_flush(dev);
 440
 441         /* Delete broadcast and local addresses since they will be recreated */
 442         ipoib_mcast_dev_down(dev);
 443
 444         ipoib_flush_paths(dev);
 445
 446         return 0;
 447 }
 448
 449 static int recvs_pending(struct net_device *dev)
 450 {
 451         struct ipoib_dev_priv *priv = netdev_priv(dev);
 452         int pending = 0;
 453         int i;
 454
 455         for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
 456                 if (priv->rx_ring[i].skb)
 457                         ++pending;
 458
 459         return pending;
 460 }
 461
 462 int ipoib_ib_dev_stop(struct net_device *dev)
 463 {
 464         struct ipoib_dev_priv *priv = netdev_priv(dev);
 465         struct ib_qp_attr qp_attr;
 466         int attr_mask;
 467         unsigned long begin;
 468         struct ipoib_buf *tx_req;
 469         int i;
 470
 471         /* Kill the existing QP and allocate a new one */
 472         qp_attr.qp_state = IB_QPS_ERR;
 473         attr_mask        = IB_QP_STATE;
 474         if (ib_modify_qp(priv->qp, &qp_attr, attr_mask))
 475                 ipoib_warn(priv, "Failed to modify QP to ERROR state\n");
 476
 477         /* Wait for all sends and receives to complete */
 478         begin = jiffies;
 479
 480         while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) {
 481                 if (time_after(jiffies, begin + 5 * HZ)) {
 482                         ipoib_warn(priv, "timing out; %d sends %d receives not completed\n",
 483                                    priv->tx_head - priv->tx_tail, recvs_pending(dev));
 484
 485                         /*
 486                          * assume the HW is wedged and just free up
 487                          * all our pending work requests.
 488                          */
 489                         while (priv->tx_tail < priv->tx_head) {
 490                                 tx_req = &priv->tx_ring[priv->tx_tail &
 491                                                         (IPOIB_TX_RING_SIZE - 1)];
 492                                 dma_unmap_single(priv->ca->dma_device,
 493                                                  pci_unmap_addr(tx_req, mapping),
 494                                                  tx_req->skb->len,
 495                                                  DMA_TO_DEVICE);
 496                                 dev_kfree_skb_any(tx_req->skb);
 497                                 ++priv->tx_tail;
 498                         }
 499
 500                         for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
 501                                 if (priv->rx_ring[i].skb) {
 502                                         dma_unmap_single(priv->ca->dma_device,
 503                                                          pci_unmap_addr(&priv->rx_ring[i],
 504                                                                         mapping),
 505                                                          IPOIB_BUF_SIZE,
 506                                                          DMA_FROM_DEVICE);
 507                                         dev_kfree_skb_any(priv->rx_ring[i].skb);
 508                                         priv->rx_ring[i].skb = NULL;
 509                                 }
 510
 511                         goto timeout;
 512                 }
 513
 514                 msleep(1);
 515         }
 516
 517         ipoib_dbg(priv, "All sends and receives done.\n");
 518
 519 timeout:
 520         qp_attr.qp_state = IB_QPS_RESET;
 521         attr_mask        = IB_QP_STATE;
 522         if (ib_modify_qp(priv->qp, &qp_attr, attr_mask))
 523                 ipoib_warn(priv, "Failed to modify QP to RESET state\n");
 524
 525         /* Wait for all AHs to be reaped */
 526         set_bit(IPOIB_STOP_REAPER, &priv->flags);
 527         cancel_delayed_work(&priv->ah_reap_task);
 528         flush_workqueue(ipoib_workqueue);
 529
 530         begin = jiffies;
 531
 532         while (!list_empty(&priv->dead_ahs)) {
 533                 __ipoib_reap_ah(dev);
 534
 535                 if (time_after(jiffies, begin + HZ)) {
 536                         ipoib_warn(priv, "timing out; will leak address handles\n");
 537                         break;
 538                 }
 539
 540                 msleep(1);
 541         }
 542
 543         return 0;
 544 }
 545
 546 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 547 {
 548         struct ipoib_dev_priv *priv = netdev_priv(dev);
 549
 550         priv->ca = ca;
 551         priv->port = port;
 552         priv->qp = NULL;
 553
 554         if (ipoib_transport_dev_init(dev, ca)) {
 555                 printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name);
 556                 return -ENODEV;
 557         }
 558
 559         if (dev->flags & IFF_UP) {
 560                 if (ipoib_ib_dev_open(dev)) {
 561                         ipoib_transport_dev_cleanup(dev);
 562                         return -ENODEV;
 563                 }
 564         }
 565
 566         return 0;
 567 }
 568
 569 void ipoib_ib_dev_flush(void *_dev)
 570 {
 571         struct net_device *dev = (struct net_device *)_dev;
 572         struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv;
 573
 574         if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
 575                 return;
 576
 577         ipoib_dbg(priv, "flushing\n");
 578
 579         ipoib_ib_dev_down(dev);
 580
 581         /*
 582          * The device could have been brought down between the start and when
 583          * we get here, don't bring it back up if it's not configured up
 584          */
 585         if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
 586                 ipoib_ib_dev_up(dev);
 587
 588         /* Flush any child interfaces too */
 589         list_for_each_entry(cpriv, &priv->child_intfs, list)
 590                 ipoib_ib_dev_flush(&cpriv->dev);
 591 }
 592
 593 void ipoib_ib_dev_cleanup(struct net_device *dev)
 594 {
 595         struct ipoib_dev_priv *priv = netdev_priv(dev);
 596
 597         ipoib_dbg(priv, "cleaning up ib_dev\n");
 598
 599         ipoib_mcast_stop_thread(dev);
 600
 601         /* Delete the broadcast address and the local address */
 602         ipoib_mcast_dev_down(dev);
 603
 604         ipoib_transport_dev_cleanup(dev);
 605 }
 606
 607 /*
 608  * Delayed P_Key Assigment Interim Support
 609  *
 610  * The following is initial implementation of delayed P_Key assigment
 611  * mechanism. It is using the same approach implemented for the multicast
 612  * group join. The single goal of this implementation is to quickly address
 613  * Bug #2507. This implementation will probably be removed when the P_Key
 614  * change async notification is available.
 615  */
 616 int ipoib_open(struct net_device *dev);
 617
 618 static void ipoib_pkey_dev_check_presence(struct net_device *dev)
 619 {
 620         struct ipoib_dev_priv *priv = netdev_priv(dev);
 621         u16 pkey_index = 0;
 622
 623         if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index))
 624                 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
 625         else
 626                 set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
 627 }
 628
 629 void ipoib_pkey_poll(void *dev_ptr)
 630 {
 631         struct net_device *dev = dev_ptr;
 632         struct ipoib_dev_priv *priv = netdev_priv(dev);
 633
 634         ipoib_pkey_dev_check_presence(dev);
 635
 636         if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
 637                 ipoib_open(dev);
 638         else {
 639                 down(&pkey_sem);
 640                 if (!test_bit(IPOIB_PKEY_STOP, &priv->flags))
 641                         queue_delayed_work(ipoib_workqueue,
 642                                            &priv->pkey_task,
 643                                            HZ);
 644                 up(&pkey_sem);
 645         }
 646 }
 647
 648 int ipoib_pkey_dev_delay_open(struct net_device *dev)
 649 {
 650         struct ipoib_dev_priv *priv = netdev_priv(dev);
 651
 652         /* Look for the interface pkey value in the IB Port P_Key table and */
 653         /* set the interface pkey assigment flag                            */
 654         ipoib_pkey_dev_check_presence(dev);
 655
 656         /* P_Key value not assigned yet - start polling */
 657         if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
 658                 down(&pkey_sem);
 659                 clear_bit(IPOIB_PKEY_STOP, &priv->flags);
 660                 queue_delayed_work(ipoib_workqueue,
 661                                    &priv->pkey_task,
 662                                    HZ);
 663                 up(&pkey_sem);
 664                 return 1;
 665         }
 666
 667         return 0;
 668 }