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.
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:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
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.
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
35 * $Id: ipoib_ib.c 1386 2004-12-27 16:23:17Z roland $
38 #include <linux/delay.h>
39 #include <linux/dma-mapping.h>
41 #include <rdma/ib_cache.h>
45 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
46 static int data_debug_level;
48 module_param(data_debug_level, int, 0644);
49 MODULE_PARM_DESC(data_debug_level,
50 "Enable data path debug tracing if > 0");
53 static DEFINE_MUTEX(pkey_mutex);
55 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
56 struct ib_pd *pd, struct ib_ah_attr *attr)
60 ah = kmalloc(sizeof *ah, GFP_KERNEL);
68 ah->ah = ib_create_ah(pd, attr);
73 ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah);
78 void ipoib_free_ah(struct kref *kref)
80 struct ipoib_ah *ah = container_of(kref, struct ipoib_ah, ref);
81 struct ipoib_dev_priv *priv = netdev_priv(ah->dev);
85 spin_lock_irqsave(&priv->lock, flags);
86 list_add_tail(&ah->list, &priv->dead_ahs);
87 spin_unlock_irqrestore(&priv->lock, flags);
90 static int ipoib_ib_post_receive(struct net_device *dev, int id)
92 struct ipoib_dev_priv *priv = netdev_priv(dev);
94 struct ib_recv_wr param;
95 struct ib_recv_wr *bad_wr;
98 list.addr = priv->rx_ring[id].mapping;
99 list.length = IPOIB_BUF_SIZE;
100 list.lkey = priv->mr->lkey;
103 param.wr_id = id | IPOIB_OP_RECV;
104 param.sg_list = &list;
107 ret = ib_post_recv(priv->qp, ¶m, &bad_wr);
109 ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret);
110 ib_dma_unmap_single(priv->ca, priv->rx_ring[id].mapping,
111 IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
112 dev_kfree_skb_any(priv->rx_ring[id].skb);
113 priv->rx_ring[id].skb = NULL;
119 static int ipoib_alloc_rx_skb(struct net_device *dev, int id)
121 struct ipoib_dev_priv *priv = netdev_priv(dev);
125 skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4);
130 * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte
131 * header. So we need 4 more bytes to get to 48 and align the
132 * IP header to a multiple of 16.
136 addr = ib_dma_map_single(priv->ca, skb->data, IPOIB_BUF_SIZE,
138 if (unlikely(ib_dma_mapping_error(priv->ca, addr))) {
139 dev_kfree_skb_any(skb);
143 priv->rx_ring[id].skb = skb;
144 priv->rx_ring[id].mapping = addr;
149 static int ipoib_ib_post_receives(struct net_device *dev)
151 struct ipoib_dev_priv *priv = netdev_priv(dev);
154 for (i = 0; i < ipoib_recvq_size; ++i) {
155 if (ipoib_alloc_rx_skb(dev, i)) {
156 ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
159 if (ipoib_ib_post_receive(dev, i)) {
160 ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i);
168 static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
170 struct ipoib_dev_priv *priv = netdev_priv(dev);
171 unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV;
175 ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n",
178 if (unlikely(wr_id >= ipoib_recvq_size)) {
179 ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n",
180 wr_id, ipoib_recvq_size);
184 skb = priv->rx_ring[wr_id].skb;
185 addr = priv->rx_ring[wr_id].mapping;
187 if (unlikely(wc->status != IB_WC_SUCCESS)) {
188 if (wc->status != IB_WC_WR_FLUSH_ERR)
189 ipoib_warn(priv, "failed recv event "
190 "(status=%d, wrid=%d vend_err %x)\n",
191 wc->status, wr_id, wc->vendor_err);
192 ib_dma_unmap_single(priv->ca, addr,
193 IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
194 dev_kfree_skb_any(skb);
195 priv->rx_ring[wr_id].skb = NULL;
200 * Drop packets that this interface sent, ie multicast packets
201 * that the HCA has replicated.
203 if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num)
207 * If we can't allocate a new RX buffer, dump
208 * this packet and reuse the old buffer.
210 if (unlikely(ipoib_alloc_rx_skb(dev, wr_id))) {
211 ++dev->stats.rx_dropped;
215 ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
216 wc->byte_len, wc->slid);
218 ib_dma_unmap_single(priv->ca, addr, IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
220 skb_put(skb, wc->byte_len);
221 skb_pull(skb, IB_GRH_BYTES);
223 skb->protocol = ((struct ipoib_header *) skb->data)->proto;
224 skb_reset_mac_header(skb);
225 skb_pull(skb, IPOIB_ENCAP_LEN);
227 dev->last_rx = jiffies;
228 ++dev->stats.rx_packets;
229 dev->stats.rx_bytes += skb->len;
232 /* XXX get correct PACKET_ type here */
233 skb->pkt_type = PACKET_HOST;
234 netif_receive_skb(skb);
237 if (unlikely(ipoib_ib_post_receive(dev, wr_id)))
238 ipoib_warn(priv, "ipoib_ib_post_receive failed "
239 "for buf %d\n", wr_id);
242 static int ipoib_dma_map_tx(struct ib_device *ca,
243 struct ipoib_tx_buf *tx_req)
245 struct sk_buff *skb = tx_req->skb;
246 u64 *mapping = tx_req->mapping;
249 mapping[0] = ib_dma_map_single(ca, skb->data, skb_headlen(skb),
251 if (unlikely(ib_dma_mapping_error(ca, mapping[0])))
254 for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) {
255 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
256 mapping[i + 1] = ib_dma_map_page(ca, frag->page,
257 frag->page_offset, frag->size,
259 if (unlikely(ib_dma_mapping_error(ca, mapping[i + 1])))
265 ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE);
268 skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
269 ib_dma_unmap_page(ca, mapping[i], frag->size, DMA_TO_DEVICE);
274 static void ipoib_dma_unmap_tx(struct ib_device *ca,
275 struct ipoib_tx_buf *tx_req)
277 struct sk_buff *skb = tx_req->skb;
278 u64 *mapping = tx_req->mapping;
281 ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE);
283 for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) {
284 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
285 ib_dma_unmap_page(ca, mapping[i + 1], frag->size,
290 static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
292 struct ipoib_dev_priv *priv = netdev_priv(dev);
293 unsigned int wr_id = wc->wr_id;
294 struct ipoib_tx_buf *tx_req;
297 ipoib_dbg_data(priv, "send completion: id %d, status: %d\n",
300 if (unlikely(wr_id >= ipoib_sendq_size)) {
301 ipoib_warn(priv, "send completion event with wrid %d (> %d)\n",
302 wr_id, ipoib_sendq_size);
306 tx_req = &priv->tx_ring[wr_id];
308 ipoib_dma_unmap_tx(priv->ca, tx_req);
310 ++dev->stats.tx_packets;
311 dev->stats.tx_bytes += tx_req->skb->len;
313 dev_kfree_skb_any(tx_req->skb);
315 spin_lock_irqsave(&priv->tx_lock, flags);
317 if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) &&
318 netif_queue_stopped(dev) &&
319 test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
320 netif_wake_queue(dev);
321 spin_unlock_irqrestore(&priv->tx_lock, flags);
323 if (wc->status != IB_WC_SUCCESS &&
324 wc->status != IB_WC_WR_FLUSH_ERR)
325 ipoib_warn(priv, "failed send event "
326 "(status=%d, wrid=%d vend_err %x)\n",
327 wc->status, wr_id, wc->vendor_err);
330 int ipoib_poll(struct napi_struct *napi, int budget)
332 struct ipoib_dev_priv *priv = container_of(napi, struct ipoib_dev_priv, napi);
333 struct net_device *dev = priv->dev;
341 while (done < budget) {
342 int max = (budget - done);
344 t = min(IPOIB_NUM_WC, max);
345 n = ib_poll_cq(priv->cq, t, priv->ibwc);
347 for (i = 0; i < n; i++) {
348 struct ib_wc *wc = priv->ibwc + i;
350 if (wc->wr_id & IPOIB_OP_RECV) {
352 if (wc->wr_id & IPOIB_OP_CM)
353 ipoib_cm_handle_rx_wc(dev, wc);
355 ipoib_ib_handle_rx_wc(dev, wc);
357 if (wc->wr_id & IPOIB_OP_CM)
358 ipoib_cm_handle_tx_wc(dev, wc);
360 ipoib_ib_handle_tx_wc(dev, wc);
369 netif_rx_complete(dev, napi);
370 if (unlikely(ib_req_notify_cq(priv->cq,
372 IB_CQ_REPORT_MISSED_EVENTS)) &&
373 netif_rx_reschedule(dev, napi))
380 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
382 struct net_device *dev = dev_ptr;
383 struct ipoib_dev_priv *priv = netdev_priv(dev);
385 netif_rx_schedule(dev, &priv->napi);
388 static inline int post_send(struct ipoib_dev_priv *priv,
390 struct ib_ah *address, u32 qpn,
391 u64 *mapping, int headlen,
395 struct ib_send_wr *bad_wr;
398 priv->tx_sge[0].addr = mapping[0];
399 priv->tx_sge[0].length = headlen;
400 for (i = 0; i < nr_frags; ++i) {
401 priv->tx_sge[i + 1].addr = mapping[i + 1];
402 priv->tx_sge[i + 1].length = frags[i].size;
404 priv->tx_wr.num_sge = nr_frags + 1;
405 priv->tx_wr.wr_id = wr_id;
406 priv->tx_wr.wr.ud.remote_qpn = qpn;
407 priv->tx_wr.wr.ud.ah = address;
409 return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr);
412 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
413 struct ipoib_ah *address, u32 qpn)
415 struct ipoib_dev_priv *priv = netdev_priv(dev);
416 struct ipoib_tx_buf *tx_req;
418 if (unlikely(skb->len > priv->mcast_mtu + IPOIB_ENCAP_LEN)) {
419 ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
420 skb->len, priv->mcast_mtu + IPOIB_ENCAP_LEN);
421 ++dev->stats.tx_dropped;
422 ++dev->stats.tx_errors;
423 ipoib_cm_skb_too_long(dev, skb, priv->mcast_mtu);
427 ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n",
428 skb->len, address, qpn);
431 * We put the skb into the tx_ring _before_ we call post_send()
432 * because it's entirely possible that the completion handler will
433 * run before we execute anything after the post_send(). That
434 * means we have to make sure everything is properly recorded and
435 * our state is consistent before we call post_send().
437 tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)];
439 if (unlikely(ipoib_dma_map_tx(priv->ca, tx_req))) {
440 ++dev->stats.tx_errors;
441 dev_kfree_skb_any(skb);
445 if (unlikely(post_send(priv, priv->tx_head & (ipoib_sendq_size - 1),
447 tx_req->mapping, skb_headlen(skb),
448 skb_shinfo(skb)->frags, skb_shinfo(skb)->nr_frags))) {
449 ipoib_warn(priv, "post_send failed\n");
450 ++dev->stats.tx_errors;
451 ipoib_dma_unmap_tx(priv->ca, tx_req);
452 dev_kfree_skb_any(skb);
454 dev->trans_start = jiffies;
456 address->last_send = priv->tx_head;
459 if (++priv->tx_outstanding == ipoib_sendq_size) {
460 ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
461 netif_stop_queue(dev);
466 static void __ipoib_reap_ah(struct net_device *dev)
468 struct ipoib_dev_priv *priv = netdev_priv(dev);
469 struct ipoib_ah *ah, *tah;
470 LIST_HEAD(remove_list);
472 spin_lock_irq(&priv->tx_lock);
473 spin_lock(&priv->lock);
474 list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list)
475 if ((int) priv->tx_tail - (int) ah->last_send >= 0) {
477 ib_destroy_ah(ah->ah);
480 spin_unlock(&priv->lock);
481 spin_unlock_irq(&priv->tx_lock);
484 void ipoib_reap_ah(struct work_struct *work)
486 struct ipoib_dev_priv *priv =
487 container_of(work, struct ipoib_dev_priv, ah_reap_task.work);
488 struct net_device *dev = priv->dev;
490 __ipoib_reap_ah(dev);
492 if (!test_bit(IPOIB_STOP_REAPER, &priv->flags))
493 queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task,
494 round_jiffies_relative(HZ));
497 int ipoib_ib_dev_open(struct net_device *dev)
499 struct ipoib_dev_priv *priv = netdev_priv(dev);
502 if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &priv->pkey_index)) {
503 ipoib_warn(priv, "P_Key 0x%04x not found\n", priv->pkey);
504 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
507 set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
509 ret = ipoib_init_qp(dev);
511 ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret);
515 ret = ipoib_ib_post_receives(dev);
517 ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret);
518 ipoib_ib_dev_stop(dev, 1);
522 ret = ipoib_cm_dev_open(dev);
524 ipoib_warn(priv, "ipoib_cm_dev_open returned %d\n", ret);
525 ipoib_ib_dev_stop(dev, 1);
529 clear_bit(IPOIB_STOP_REAPER, &priv->flags);
530 queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task,
531 round_jiffies_relative(HZ));
533 set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags);
538 static void ipoib_pkey_dev_check_presence(struct net_device *dev)
540 struct ipoib_dev_priv *priv = netdev_priv(dev);
543 if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index))
544 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
546 set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
549 int ipoib_ib_dev_up(struct net_device *dev)
551 struct ipoib_dev_priv *priv = netdev_priv(dev);
553 ipoib_pkey_dev_check_presence(dev);
555 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
556 ipoib_dbg(priv, "PKEY is not assigned.\n");
560 set_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
562 return ipoib_mcast_start_thread(dev);
565 int ipoib_ib_dev_down(struct net_device *dev, int flush)
567 struct ipoib_dev_priv *priv = netdev_priv(dev);
569 ipoib_dbg(priv, "downing ib_dev\n");
571 clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
572 netif_carrier_off(dev);
574 /* Shutdown the P_Key thread if still active */
575 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
576 mutex_lock(&pkey_mutex);
577 set_bit(IPOIB_PKEY_STOP, &priv->flags);
578 cancel_delayed_work(&priv->pkey_poll_task);
579 mutex_unlock(&pkey_mutex);
581 flush_workqueue(ipoib_workqueue);
584 ipoib_mcast_stop_thread(dev, flush);
585 ipoib_mcast_dev_flush(dev);
587 ipoib_flush_paths(dev);
592 static int recvs_pending(struct net_device *dev)
594 struct ipoib_dev_priv *priv = netdev_priv(dev);
598 for (i = 0; i < ipoib_recvq_size; ++i)
599 if (priv->rx_ring[i].skb)
605 void ipoib_drain_cq(struct net_device *dev)
607 struct ipoib_dev_priv *priv = netdev_priv(dev);
610 n = ib_poll_cq(priv->cq, IPOIB_NUM_WC, priv->ibwc);
611 for (i = 0; i < n; ++i) {
613 * Convert any successful completions to flush
614 * errors to avoid passing packets up the
615 * stack after bringing the device down.
617 if (priv->ibwc[i].status == IB_WC_SUCCESS)
618 priv->ibwc[i].status = IB_WC_WR_FLUSH_ERR;
620 if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) {
621 if (priv->ibwc[i].wr_id & IPOIB_OP_CM)
622 ipoib_cm_handle_rx_wc(dev, priv->ibwc + i);
624 ipoib_ib_handle_rx_wc(dev, priv->ibwc + i);
626 if (priv->ibwc[i].wr_id & IPOIB_OP_CM)
627 ipoib_cm_handle_tx_wc(dev, priv->ibwc + i);
629 ipoib_ib_handle_tx_wc(dev, priv->ibwc + i);
632 } while (n == IPOIB_NUM_WC);
635 int ipoib_ib_dev_stop(struct net_device *dev, int flush)
637 struct ipoib_dev_priv *priv = netdev_priv(dev);
638 struct ib_qp_attr qp_attr;
640 struct ipoib_tx_buf *tx_req;
643 clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags);
645 ipoib_cm_dev_stop(dev);
648 * Move our QP to the error state and then reinitialize in
649 * when all work requests have completed or have been flushed.
651 qp_attr.qp_state = IB_QPS_ERR;
652 if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
653 ipoib_warn(priv, "Failed to modify QP to ERROR state\n");
655 /* Wait for all sends and receives to complete */
658 while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) {
659 if (time_after(jiffies, begin + 5 * HZ)) {
660 ipoib_warn(priv, "timing out; %d sends %d receives not completed\n",
661 priv->tx_head - priv->tx_tail, recvs_pending(dev));
664 * assume the HW is wedged and just free up
665 * all our pending work requests.
667 while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
668 tx_req = &priv->tx_ring[priv->tx_tail &
669 (ipoib_sendq_size - 1)];
670 ipoib_dma_unmap_tx(priv->ca, tx_req);
671 dev_kfree_skb_any(tx_req->skb);
673 --priv->tx_outstanding;
676 for (i = 0; i < ipoib_recvq_size; ++i) {
677 struct ipoib_rx_buf *rx_req;
679 rx_req = &priv->rx_ring[i];
682 ib_dma_unmap_single(priv->ca,
686 dev_kfree_skb_any(rx_req->skb);
698 ipoib_dbg(priv, "All sends and receives done.\n");
701 qp_attr.qp_state = IB_QPS_RESET;
702 if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
703 ipoib_warn(priv, "Failed to modify QP to RESET state\n");
705 /* Wait for all AHs to be reaped */
706 set_bit(IPOIB_STOP_REAPER, &priv->flags);
707 cancel_delayed_work(&priv->ah_reap_task);
709 flush_workqueue(ipoib_workqueue);
713 while (!list_empty(&priv->dead_ahs)) {
714 __ipoib_reap_ah(dev);
716 if (time_after(jiffies, begin + HZ)) {
717 ipoib_warn(priv, "timing out; will leak address handles\n");
724 ib_req_notify_cq(priv->cq, IB_CQ_NEXT_COMP);
729 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
731 struct ipoib_dev_priv *priv = netdev_priv(dev);
737 if (ipoib_transport_dev_init(dev, ca)) {
738 printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name);
742 if (dev->flags & IFF_UP) {
743 if (ipoib_ib_dev_open(dev)) {
744 ipoib_transport_dev_cleanup(dev);
752 static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, int pkey_event)
754 struct ipoib_dev_priv *cpriv;
755 struct net_device *dev = priv->dev;
758 mutex_lock(&priv->vlan_mutex);
761 * Flush any child interfaces too -- they might be up even if
762 * the parent is down.
764 list_for_each_entry(cpriv, &priv->child_intfs, list)
765 __ipoib_ib_dev_flush(cpriv, pkey_event);
767 mutex_unlock(&priv->vlan_mutex);
769 if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) {
770 ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n");
774 if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
775 ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n");
780 if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &new_index)) {
781 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
782 ipoib_ib_dev_down(dev, 0);
783 ipoib_ib_dev_stop(dev, 0);
784 ipoib_pkey_dev_delay_open(dev);
787 set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
789 /* restart QP only if P_Key index is changed */
790 if (new_index == priv->pkey_index) {
791 ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n");
794 priv->pkey_index = new_index;
797 ipoib_dbg(priv, "flushing\n");
799 ipoib_ib_dev_down(dev, 0);
802 ipoib_ib_dev_stop(dev, 0);
803 ipoib_ib_dev_open(dev);
807 * The device could have been brought down between the start and when
808 * we get here, don't bring it back up if it's not configured up
810 if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
811 ipoib_ib_dev_up(dev);
812 ipoib_mcast_restart_task(&priv->restart_task);
816 void ipoib_ib_dev_flush(struct work_struct *work)
818 struct ipoib_dev_priv *priv =
819 container_of(work, struct ipoib_dev_priv, flush_task);
821 ipoib_dbg(priv, "Flushing %s\n", priv->dev->name);
822 __ipoib_ib_dev_flush(priv, 0);
825 void ipoib_pkey_event(struct work_struct *work)
827 struct ipoib_dev_priv *priv =
828 container_of(work, struct ipoib_dev_priv, pkey_event_task);
830 ipoib_dbg(priv, "Flushing %s and restarting its QP\n", priv->dev->name);
831 __ipoib_ib_dev_flush(priv, 1);
834 void ipoib_ib_dev_cleanup(struct net_device *dev)
836 struct ipoib_dev_priv *priv = netdev_priv(dev);
838 ipoib_dbg(priv, "cleaning up ib_dev\n");
840 ipoib_mcast_stop_thread(dev, 1);
841 ipoib_mcast_dev_flush(dev);
843 ipoib_transport_dev_cleanup(dev);
847 * Delayed P_Key Assigment Interim Support
849 * The following is initial implementation of delayed P_Key assigment
850 * mechanism. It is using the same approach implemented for the multicast
851 * group join. The single goal of this implementation is to quickly address
852 * Bug #2507. This implementation will probably be removed when the P_Key
853 * change async notification is available.
856 void ipoib_pkey_poll(struct work_struct *work)
858 struct ipoib_dev_priv *priv =
859 container_of(work, struct ipoib_dev_priv, pkey_poll_task.work);
860 struct net_device *dev = priv->dev;
862 ipoib_pkey_dev_check_presence(dev);
864 if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
867 mutex_lock(&pkey_mutex);
868 if (!test_bit(IPOIB_PKEY_STOP, &priv->flags))
869 queue_delayed_work(ipoib_workqueue,
870 &priv->pkey_poll_task,
872 mutex_unlock(&pkey_mutex);
876 int ipoib_pkey_dev_delay_open(struct net_device *dev)
878 struct ipoib_dev_priv *priv = netdev_priv(dev);
880 /* Look for the interface pkey value in the IB Port P_Key table and */
881 /* set the interface pkey assigment flag */
882 ipoib_pkey_dev_check_presence(dev);
884 /* P_Key value not assigned yet - start polling */
885 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
886 mutex_lock(&pkey_mutex);
887 clear_bit(IPOIB_PKEY_STOP, &priv->flags);
888 queue_delayed_work(ipoib_workqueue,
889 &priv->pkey_poll_task,
891 mutex_unlock(&pkey_mutex);