1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
5  * Copyright (c) 2004, 2005 Voltaire, Inc. All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  *
35  * $Id: ipoib_ib.c 1386 2004-12-27 16:23:17Z roland $
36  */
37 
38 #include <linux/delay.h>
39 #include <linux/dma-mapping.h>
40 
41 #include <rdma/ib_cache.h>
42 
43 #include "ipoib.h"
44 
45 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
46 static int data_debug_level;
47 
48 module_param(data_debug_level, int, 0644);
49 MODULE_PARM_DESC(data_debug_level,
50 		 "Enable data path debug tracing if > 0");
51 #endif
52 
53 #define	IPOIB_OP_RECV	(1ul << 31)
54 
55 static DECLARE_MUTEX(pkey_sem);
56 
57 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
58 				 struct ib_pd *pd, struct ib_ah_attr *attr)
59 {
60 	struct ipoib_ah *ah;
61 
62 	ah = kmalloc(sizeof *ah, GFP_KERNEL);
63 	if (!ah)
64 		return NULL;
65 
66 	ah->dev       = dev;
67 	ah->last_send = 0;
68 	kref_init(&ah->ref);
69 
70 	ah->ah = ib_create_ah(pd, attr);
71 	if (IS_ERR(ah->ah)) {
72 		kfree(ah);
73 		ah = NULL;
74 	} else
75 		ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah);
76 
77 	return ah;
78 }
79 
80 void ipoib_free_ah(struct kref *kref)
81 {
82 	struct ipoib_ah *ah = container_of(kref, struct ipoib_ah, ref);
83 	struct ipoib_dev_priv *priv = netdev_priv(ah->dev);
84 
85 	unsigned long flags;
86 
87 	if ((int) priv->tx_tail - (int) ah->last_send >= 0) {
88 		ipoib_dbg(priv, "Freeing ah %p\n", ah->ah);
89 		ib_destroy_ah(ah->ah);
90 		kfree(ah);
91 	} else {
92 		spin_lock_irqsave(&priv->lock, flags);
93 		list_add_tail(&ah->list, &priv->dead_ahs);
94 		spin_unlock_irqrestore(&priv->lock, flags);
95 	}
96 }
97 
98 static int ipoib_ib_post_receive(struct net_device *dev, int id)
99 {
100 	struct ipoib_dev_priv *priv = netdev_priv(dev);
101 	struct ib_sge list;
102 	struct ib_recv_wr param;
103 	struct ib_recv_wr *bad_wr;
104 	int ret;
105 
106 	list.addr     = priv->rx_ring[id].mapping;
107 	list.length   = IPOIB_BUF_SIZE;
108 	list.lkey     = priv->mr->lkey;
109 
110 	param.next    = NULL;
111 	param.wr_id   = id | IPOIB_OP_RECV;
112 	param.sg_list = &list;
113 	param.num_sge = 1;
114 
115 	ret = ib_post_recv(priv->qp, &param, &bad_wr);
116 	if (unlikely(ret)) {
117 		ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret);
118 		dma_unmap_single(priv->ca->dma_device,
119 				 priv->rx_ring[id].mapping,
120 				 IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
121 		dev_kfree_skb_any(priv->rx_ring[id].skb);
122 		priv->rx_ring[id].skb = NULL;
123 	}
124 
125 	return ret;
126 }
127 
128 static int ipoib_alloc_rx_skb(struct net_device *dev, int id)
129 {
130 	struct ipoib_dev_priv *priv = netdev_priv(dev);
131 	struct sk_buff *skb;
132 	dma_addr_t addr;
133 
134 	skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4);
135 	if (!skb)
136 		return -ENOMEM;
137 
138 	/*
139 	 * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte
140 	 * header.  So we need 4 more bytes to get to 48 and align the
141 	 * IP header to a multiple of 16.
142 	 */
143 	skb_reserve(skb, 4);
144 
145 	addr = dma_map_single(priv->ca->dma_device,
146 			      skb->data, IPOIB_BUF_SIZE,
147 			      DMA_FROM_DEVICE);
148 	if (unlikely(dma_mapping_error(addr))) {
149 		dev_kfree_skb_any(skb);
150 		return -EIO;
151 	}
152 
153 	priv->rx_ring[id].skb     = skb;
154 	priv->rx_ring[id].mapping = addr;
155 
156 	return 0;
157 }
158 
159 static int ipoib_ib_post_receives(struct net_device *dev)
160 {
161 	struct ipoib_dev_priv *priv = netdev_priv(dev);
162 	int i;
163 
164 	for (i = 0; i < IPOIB_RX_RING_SIZE; ++i) {
165 		if (ipoib_alloc_rx_skb(dev, i)) {
166 			ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
167 			return -ENOMEM;
168 		}
169 		if (ipoib_ib_post_receive(dev, i)) {
170 			ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i);
171 			return -EIO;
172 		}
173 	}
174 
175 	return 0;
176 }
177 
178 static void ipoib_ib_handle_wc(struct net_device *dev,
179 			       struct ib_wc *wc)
180 {
181 	struct ipoib_dev_priv *priv = netdev_priv(dev);
182 	unsigned int wr_id = wc->wr_id;
183 
184 	ipoib_dbg_data(priv, "called: id %d, op %d, status: %d\n",
185 		       wr_id, wc->opcode, wc->status);
186 
187 	if (wr_id & IPOIB_OP_RECV) {
188 		wr_id &= ~IPOIB_OP_RECV;
189 
190 		if (wr_id < IPOIB_RX_RING_SIZE) {
191 			struct sk_buff *skb  = priv->rx_ring[wr_id].skb;
192 			dma_addr_t      addr = priv->rx_ring[wr_id].mapping;
193 
194 			if (unlikely(wc->status != IB_WC_SUCCESS)) {
195 				if (wc->status != IB_WC_WR_FLUSH_ERR)
196 					ipoib_warn(priv, "failed recv event "
197 						   "(status=%d, wrid=%d vend_err %x)\n",
198 						   wc->status, wr_id, wc->vendor_err);
199 				dma_unmap_single(priv->ca->dma_device, addr,
200 						 IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
201 				dev_kfree_skb_any(skb);
202 				priv->rx_ring[wr_id].skb = NULL;
203 				return;
204 			}
205 
206 			/*
207 			 * If we can't allocate a new RX buffer, dump
208 			 * this packet and reuse the old buffer.
209 			 */
210 			if (unlikely(ipoib_alloc_rx_skb(dev, wr_id))) {
211 				++priv->stats.rx_dropped;
212 				goto repost;
213 			}
214 
215 			ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
216 				       wc->byte_len, wc->slid);
217 
218 			dma_unmap_single(priv->ca->dma_device, addr,
219 					 IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
220 
221 			skb_put(skb, wc->byte_len);
222 			skb_pull(skb, IB_GRH_BYTES);
223 
224 			if (wc->slid != priv->local_lid ||
225 			    wc->src_qp != priv->qp->qp_num) {
226 				skb->protocol = ((struct ipoib_header *) skb->data)->proto;
227 				skb->mac.raw = skb->data;
228 				skb_pull(skb, IPOIB_ENCAP_LEN);
229 
230 				dev->last_rx = jiffies;
231 				++priv->stats.rx_packets;
232 				priv->stats.rx_bytes += skb->len;
233 
234 				skb->dev = dev;
235 				/* XXX get correct PACKET_ type here */
236 				skb->pkt_type = PACKET_HOST;
237 				netif_rx_ni(skb);
238 			} else {
239 				ipoib_dbg_data(priv, "dropping loopback packet\n");
240 				dev_kfree_skb_any(skb);
241 			}
242 
243 		repost:
244 			if (unlikely(ipoib_ib_post_receive(dev, wr_id)))
245 				ipoib_warn(priv, "ipoib_ib_post_receive failed "
246 					   "for buf %d\n", wr_id);
247 		} else
248 			ipoib_warn(priv, "completion event with wrid %d\n",
249 				   wr_id);
250 
251 	} else {
252 		struct ipoib_tx_buf *tx_req;
253 		unsigned long flags;
254 
255 		if (wr_id >= IPOIB_TX_RING_SIZE) {
256 			ipoib_warn(priv, "completion event with wrid %d (> %d)\n",
257 				   wr_id, IPOIB_TX_RING_SIZE);
258 			return;
259 		}
260 
261 		ipoib_dbg_data(priv, "send complete, wrid %d\n", wr_id);
262 
263 		tx_req = &priv->tx_ring[wr_id];
264 
265 		dma_unmap_single(priv->ca->dma_device,
266 				 pci_unmap_addr(tx_req, mapping),
267 				 tx_req->skb->len,
268 				 DMA_TO_DEVICE);
269 
270 		++priv->stats.tx_packets;
271 		priv->stats.tx_bytes += tx_req->skb->len;
272 
273 		dev_kfree_skb_any(tx_req->skb);
274 
275 		spin_lock_irqsave(&priv->tx_lock, flags);
276 		++priv->tx_tail;
277 		if (netif_queue_stopped(dev) &&
278 		    priv->tx_head - priv->tx_tail <= IPOIB_TX_RING_SIZE / 2)
279 			netif_wake_queue(dev);
280 		spin_unlock_irqrestore(&priv->tx_lock, flags);
281 
282 		if (wc->status != IB_WC_SUCCESS &&
283 		    wc->status != IB_WC_WR_FLUSH_ERR)
284 			ipoib_warn(priv, "failed send event "
285 				   "(status=%d, wrid=%d vend_err %x)\n",
286 				   wc->status, wr_id, wc->vendor_err);
287 	}
288 }
289 
290 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
291 {
292 	struct net_device *dev = (struct net_device *) dev_ptr;
293 	struct ipoib_dev_priv *priv = netdev_priv(dev);
294 	int n, i;
295 
296 	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
297 	do {
298 		n = ib_poll_cq(cq, IPOIB_NUM_WC, priv->ibwc);
299 		for (i = 0; i < n; ++i)
300 			ipoib_ib_handle_wc(dev, priv->ibwc + i);
301 	} while (n == IPOIB_NUM_WC);
302 }
303 
304 static inline int post_send(struct ipoib_dev_priv *priv,
305 			    unsigned int wr_id,
306 			    struct ib_ah *address, u32 qpn,
307 			    dma_addr_t addr, int len)
308 {
309 	struct ib_send_wr *bad_wr;
310 
311 	priv->tx_sge.addr             = addr;
312 	priv->tx_sge.length           = len;
313 
314 	priv->tx_wr.wr_id 	      = wr_id;
315 	priv->tx_wr.wr.ud.remote_qpn  = qpn;
316 	priv->tx_wr.wr.ud.ah 	      = address;
317 
318 	return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr);
319 }
320 
321 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
322 		struct ipoib_ah *address, u32 qpn)
323 {
324 	struct ipoib_dev_priv *priv = netdev_priv(dev);
325 	struct ipoib_tx_buf *tx_req;
326 	dma_addr_t addr;
327 
328 	if (skb->len > dev->mtu + INFINIBAND_ALEN) {
329 		ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
330 			   skb->len, dev->mtu + INFINIBAND_ALEN);
331 		++priv->stats.tx_dropped;
332 		++priv->stats.tx_errors;
333 		dev_kfree_skb_any(skb);
334 		return;
335 	}
336 
337 	ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n",
338 		       skb->len, address, qpn);
339 
340 	/*
341 	 * We put the skb into the tx_ring _before_ we call post_send()
342 	 * because it's entirely possible that the completion handler will
343 	 * run before we execute anything after the post_send().  That
344 	 * means we have to make sure everything is properly recorded and
345 	 * our state is consistent before we call post_send().
346 	 */
347 	tx_req = &priv->tx_ring[priv->tx_head & (IPOIB_TX_RING_SIZE - 1)];
348 	tx_req->skb = skb;
349 	addr = dma_map_single(priv->ca->dma_device, skb->data, skb->len,
350 			      DMA_TO_DEVICE);
351 	pci_unmap_addr_set(tx_req, mapping, addr);
352 
353 	if (unlikely(post_send(priv, priv->tx_head & (IPOIB_TX_RING_SIZE - 1),
354 			       address->ah, qpn, addr, skb->len))) {
355 		ipoib_warn(priv, "post_send failed\n");
356 		++priv->stats.tx_errors;
357 		dma_unmap_single(priv->ca->dma_device, addr, skb->len,
358 				 DMA_TO_DEVICE);
359 		dev_kfree_skb_any(skb);
360 	} else {
361 		dev->trans_start = jiffies;
362 
363 		address->last_send = priv->tx_head;
364 		++priv->tx_head;
365 
366 		if (priv->tx_head - priv->tx_tail == IPOIB_TX_RING_SIZE) {
367 			ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
368 			netif_stop_queue(dev);
369 		}
370 	}
371 }
372 
373 static void __ipoib_reap_ah(struct net_device *dev)
374 {
375 	struct ipoib_dev_priv *priv = netdev_priv(dev);
376 	struct ipoib_ah *ah, *tah;
377 	LIST_HEAD(remove_list);
378 
379 	spin_lock_irq(&priv->lock);
380 	list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list)
381 		if ((int) priv->tx_tail - (int) ah->last_send >= 0) {
382 			list_del(&ah->list);
383 			list_add_tail(&ah->list, &remove_list);
384 		}
385 	spin_unlock_irq(&priv->lock);
386 
387 	list_for_each_entry_safe(ah, tah, &remove_list, list) {
388 		ipoib_dbg(priv, "Reaping ah %p\n", ah->ah);
389 		ib_destroy_ah(ah->ah);
390 		kfree(ah);
391 	}
392 }
393 
394 void ipoib_reap_ah(void *dev_ptr)
395 {
396 	struct net_device *dev = dev_ptr;
397 	struct ipoib_dev_priv *priv = netdev_priv(dev);
398 
399 	__ipoib_reap_ah(dev);
400 
401 	if (!test_bit(IPOIB_STOP_REAPER, &priv->flags))
402 		queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
403 }
404 
405 int ipoib_ib_dev_open(struct net_device *dev)
406 {
407 	struct ipoib_dev_priv *priv = netdev_priv(dev);
408 	int ret;
409 
410 	ret = ipoib_init_qp(dev);
411 	if (ret) {
412 		ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret);
413 		return -1;
414 	}
415 
416 	ret = ipoib_ib_post_receives(dev);
417 	if (ret) {
418 		ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret);
419 		return -1;
420 	}
421 
422 	clear_bit(IPOIB_STOP_REAPER, &priv->flags);
423 	queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
424 
425 	return 0;
426 }
427 
428 int ipoib_ib_dev_up(struct net_device *dev)
429 {
430 	struct ipoib_dev_priv *priv = netdev_priv(dev);
431 
432 	set_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
433 
434 	return ipoib_mcast_start_thread(dev);
435 }
436 
437 int ipoib_ib_dev_down(struct net_device *dev)
438 {
439 	struct ipoib_dev_priv *priv = netdev_priv(dev);
440 
441 	ipoib_dbg(priv, "downing ib_dev\n");
442 
443 	clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
444 	netif_carrier_off(dev);
445 
446 	/* Shutdown the P_Key thread if still active */
447 	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
448 		down(&pkey_sem);
449 		set_bit(IPOIB_PKEY_STOP, &priv->flags);
450 		cancel_delayed_work(&priv->pkey_task);
451 		up(&pkey_sem);
452 		flush_workqueue(ipoib_workqueue);
453 	}
454 
455 	ipoib_mcast_stop_thread(dev, 1);
456 
457 	/*
458 	 * Flush the multicast groups first so we stop any multicast joins. The
459 	 * completion thread may have already died and we may deadlock waiting
460 	 * for the completion thread to finish some multicast joins.
461 	 */
462 	ipoib_mcast_dev_flush(dev);
463 
464 	/* Delete broadcast and local addresses since they will be recreated */
465 	ipoib_mcast_dev_down(dev);
466 
467 	ipoib_flush_paths(dev);
468 
469 	return 0;
470 }
471 
472 static int recvs_pending(struct net_device *dev)
473 {
474 	struct ipoib_dev_priv *priv = netdev_priv(dev);
475 	int pending = 0;
476 	int i;
477 
478 	for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
479 		if (priv->rx_ring[i].skb)
480 			++pending;
481 
482 	return pending;
483 }
484 
485 int ipoib_ib_dev_stop(struct net_device *dev)
486 {
487 	struct ipoib_dev_priv *priv = netdev_priv(dev);
488 	struct ib_qp_attr qp_attr;
489 	unsigned long begin;
490 	struct ipoib_tx_buf *tx_req;
491 	int i;
492 
493 	/*
494 	 * Move our QP to the error state and then reinitialize in
495 	 * when all work requests have completed or have been flushed.
496 	 */
497 	qp_attr.qp_state = IB_QPS_ERR;
498 	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
499 		ipoib_warn(priv, "Failed to modify QP to ERROR state\n");
500 
501 	/* Wait for all sends and receives to complete */
502 	begin = jiffies;
503 
504 	while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) {
505 		if (time_after(jiffies, begin + 5 * HZ)) {
506 			ipoib_warn(priv, "timing out; %d sends %d receives not completed\n",
507 				   priv->tx_head - priv->tx_tail, recvs_pending(dev));
508 
509 			/*
510 			 * assume the HW is wedged and just free up
511 			 * all our pending work requests.
512 			 */
513 			while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
514 				tx_req = &priv->tx_ring[priv->tx_tail &
515 							(IPOIB_TX_RING_SIZE - 1)];
516 				dma_unmap_single(priv->ca->dma_device,
517 						 pci_unmap_addr(tx_req, mapping),
518 						 tx_req->skb->len,
519 						 DMA_TO_DEVICE);
520 				dev_kfree_skb_any(tx_req->skb);
521 				++priv->tx_tail;
522 			}
523 
524 			for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
525 				if (priv->rx_ring[i].skb) {
526 					dma_unmap_single(priv->ca->dma_device,
527 							 pci_unmap_addr(&priv->rx_ring[i],
528 									mapping),
529 							 IPOIB_BUF_SIZE,
530 							 DMA_FROM_DEVICE);
531 					dev_kfree_skb_any(priv->rx_ring[i].skb);
532 					priv->rx_ring[i].skb = NULL;
533 				}
534 
535 			goto timeout;
536 		}
537 
538 		msleep(1);
539 	}
540 
541 	ipoib_dbg(priv, "All sends and receives done.\n");
542 
543 timeout:
544 	qp_attr.qp_state = IB_QPS_RESET;
545 	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
546 		ipoib_warn(priv, "Failed to modify QP to RESET state\n");
547 
548 	/* Wait for all AHs to be reaped */
549 	set_bit(IPOIB_STOP_REAPER, &priv->flags);
550 	cancel_delayed_work(&priv->ah_reap_task);
551 	flush_workqueue(ipoib_workqueue);
552 
553 	begin = jiffies;
554 
555 	while (!list_empty(&priv->dead_ahs)) {
556 		__ipoib_reap_ah(dev);
557 
558 		if (time_after(jiffies, begin + HZ)) {
559 			ipoib_warn(priv, "timing out; will leak address handles\n");
560 			break;
561 		}
562 
563 		msleep(1);
564 	}
565 
566 	return 0;
567 }
568 
569 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
570 {
571 	struct ipoib_dev_priv *priv = netdev_priv(dev);
572 
573 	priv->ca = ca;
574 	priv->port = port;
575 	priv->qp = NULL;
576 
577 	if (ipoib_transport_dev_init(dev, ca)) {
578 		printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name);
579 		return -ENODEV;
580 	}
581 
582 	if (dev->flags & IFF_UP) {
583 		if (ipoib_ib_dev_open(dev)) {
584 			ipoib_transport_dev_cleanup(dev);
585 			return -ENODEV;
586 		}
587 	}
588 
589 	return 0;
590 }
591 
592 void ipoib_ib_dev_flush(void *_dev)
593 {
594 	struct net_device *dev = (struct net_device *)_dev;
595 	struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv;
596 
597 	if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
598 		return;
599 
600 	ipoib_dbg(priv, "flushing\n");
601 
602 	ipoib_ib_dev_down(dev);
603 
604 	/*
605 	 * The device could have been brought down between the start and when
606 	 * we get here, don't bring it back up if it's not configured up
607 	 */
608 	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
609 		ipoib_ib_dev_up(dev);
610 
611 	down(&priv->vlan_mutex);
612 
613 	/* Flush any child interfaces too */
614 	list_for_each_entry(cpriv, &priv->child_intfs, list)
615 		ipoib_ib_dev_flush(&cpriv->dev);
616 
617 	up(&priv->vlan_mutex);
618 }
619 
620 void ipoib_ib_dev_cleanup(struct net_device *dev)
621 {
622 	struct ipoib_dev_priv *priv = netdev_priv(dev);
623 
624 	ipoib_dbg(priv, "cleaning up ib_dev\n");
625 
626 	ipoib_mcast_stop_thread(dev, 1);
627 
628 	/* Delete the broadcast address and the local address */
629 	ipoib_mcast_dev_down(dev);
630 
631 	ipoib_transport_dev_cleanup(dev);
632 }
633 
634 /*
635  * Delayed P_Key Assigment Interim Support
636  *
637  * The following is initial implementation of delayed P_Key assigment
638  * mechanism. It is using the same approach implemented for the multicast
639  * group join. The single goal of this implementation is to quickly address
640  * Bug #2507. This implementation will probably be removed when the P_Key
641  * change async notification is available.
642  */
643 
644 static void ipoib_pkey_dev_check_presence(struct net_device *dev)
645 {
646 	struct ipoib_dev_priv *priv = netdev_priv(dev);
647 	u16 pkey_index = 0;
648 
649 	if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index))
650 		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
651 	else
652 		set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
653 }
654 
655 void ipoib_pkey_poll(void *dev_ptr)
656 {
657 	struct net_device *dev = dev_ptr;
658 	struct ipoib_dev_priv *priv = netdev_priv(dev);
659 
660 	ipoib_pkey_dev_check_presence(dev);
661 
662 	if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
663 		ipoib_open(dev);
664 	else {
665 		down(&pkey_sem);
666 		if (!test_bit(IPOIB_PKEY_STOP, &priv->flags))
667 			queue_delayed_work(ipoib_workqueue,
668 					   &priv->pkey_task,
669 					   HZ);
670 		up(&pkey_sem);
671 	}
672 }
673 
674 int ipoib_pkey_dev_delay_open(struct net_device *dev)
675 {
676 	struct ipoib_dev_priv *priv = netdev_priv(dev);
677 
678 	/* Look for the interface pkey value in the IB Port P_Key table and */
679 	/* set the interface pkey assigment flag                            */
680 	ipoib_pkey_dev_check_presence(dev);
681 
682 	/* P_Key value not assigned yet - start polling */
683 	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
684 		down(&pkey_sem);
685 		clear_bit(IPOIB_PKEY_STOP, &priv->flags);
686 		queue_delayed_work(ipoib_workqueue,
687 				   &priv->pkey_task,
688 				   HZ);
689 		up(&pkey_sem);
690 		return 1;
691 	}
692 
693 	return 0;
694 }
695