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 ++priv
->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 ++priv
->stats
.rx_packets
;
229 priv
->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 void ipoib_ib_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
244 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
245 unsigned int wr_id
= wc
->wr_id
;
246 struct ipoib_tx_buf
*tx_req
;
249 ipoib_dbg_data(priv
, "send completion: id %d, status: %d\n",
252 if (unlikely(wr_id
>= ipoib_sendq_size
)) {
253 ipoib_warn(priv
, "send completion event with wrid %d (> %d)\n",
254 wr_id
, ipoib_sendq_size
);
258 tx_req
= &priv
->tx_ring
[wr_id
];
260 ib_dma_unmap_single(priv
->ca
, tx_req
->mapping
,
261 tx_req
->skb
->len
, DMA_TO_DEVICE
);
263 ++priv
->stats
.tx_packets
;
264 priv
->stats
.tx_bytes
+= tx_req
->skb
->len
;
266 dev_kfree_skb_any(tx_req
->skb
);
268 spin_lock_irqsave(&priv
->tx_lock
, flags
);
270 if (unlikely(test_bit(IPOIB_FLAG_NETIF_STOPPED
, &priv
->flags
)) &&
271 priv
->tx_head
- priv
->tx_tail
<= ipoib_sendq_size
>> 1) {
272 clear_bit(IPOIB_FLAG_NETIF_STOPPED
, &priv
->flags
);
273 netif_wake_queue(dev
);
275 spin_unlock_irqrestore(&priv
->tx_lock
, flags
);
277 if (wc
->status
!= IB_WC_SUCCESS
&&
278 wc
->status
!= IB_WC_WR_FLUSH_ERR
)
279 ipoib_warn(priv
, "failed send event "
280 "(status=%d, wrid=%d vend_err %x)\n",
281 wc
->status
, wr_id
, wc
->vendor_err
);
284 int ipoib_poll(struct napi_struct
*napi
, int budget
)
286 struct ipoib_dev_priv
*priv
= container_of(napi
, struct ipoib_dev_priv
, napi
);
287 struct net_device
*dev
= priv
->dev
;
295 while (done
< budget
) {
296 int max
= (budget
- done
);
298 t
= min(IPOIB_NUM_WC
, max
);
299 n
= ib_poll_cq(priv
->cq
, t
, priv
->ibwc
);
301 for (i
= 0; i
< n
; i
++) {
302 struct ib_wc
*wc
= priv
->ibwc
+ i
;
304 if (wc
->wr_id
& IPOIB_CM_OP_SRQ
) {
306 ipoib_cm_handle_rx_wc(dev
, wc
);
307 } else if (wc
->wr_id
& IPOIB_OP_RECV
) {
309 ipoib_ib_handle_rx_wc(dev
, wc
);
311 ipoib_ib_handle_tx_wc(dev
, wc
);
319 netif_rx_complete(dev
, napi
);
320 if (unlikely(ib_req_notify_cq(priv
->cq
,
322 IB_CQ_REPORT_MISSED_EVENTS
)) &&
323 netif_rx_reschedule(dev
, napi
))
330 void ipoib_ib_completion(struct ib_cq
*cq
, void *dev_ptr
)
332 struct net_device
*dev
= dev_ptr
;
333 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
335 netif_rx_schedule(dev
, &priv
->napi
);
338 static inline int post_send(struct ipoib_dev_priv
*priv
,
340 struct ib_ah
*address
, u32 qpn
,
343 struct ib_send_wr
*bad_wr
;
345 priv
->tx_sge
.addr
= addr
;
346 priv
->tx_sge
.length
= len
;
348 priv
->tx_wr
.wr_id
= wr_id
;
349 priv
->tx_wr
.wr
.ud
.remote_qpn
= qpn
;
350 priv
->tx_wr
.wr
.ud
.ah
= address
;
352 return ib_post_send(priv
->qp
, &priv
->tx_wr
, &bad_wr
);
355 void ipoib_send(struct net_device
*dev
, struct sk_buff
*skb
,
356 struct ipoib_ah
*address
, u32 qpn
)
358 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
359 struct ipoib_tx_buf
*tx_req
;
362 if (unlikely(skb
->len
> priv
->mcast_mtu
+ IPOIB_ENCAP_LEN
)) {
363 ipoib_warn(priv
, "packet len %d (> %d) too long to send, dropping\n",
364 skb
->len
, priv
->mcast_mtu
+ IPOIB_ENCAP_LEN
);
365 ++priv
->stats
.tx_dropped
;
366 ++priv
->stats
.tx_errors
;
367 ipoib_cm_skb_too_long(dev
, skb
, priv
->mcast_mtu
);
371 ipoib_dbg_data(priv
, "sending packet, length=%d address=%p qpn=0x%06x\n",
372 skb
->len
, address
, qpn
);
375 * We put the skb into the tx_ring _before_ we call post_send()
376 * because it's entirely possible that the completion handler will
377 * run before we execute anything after the post_send(). That
378 * means we have to make sure everything is properly recorded and
379 * our state is consistent before we call post_send().
381 tx_req
= &priv
->tx_ring
[priv
->tx_head
& (ipoib_sendq_size
- 1)];
383 addr
= ib_dma_map_single(priv
->ca
, skb
->data
, skb
->len
,
385 if (unlikely(ib_dma_mapping_error(priv
->ca
, addr
))) {
386 ++priv
->stats
.tx_errors
;
387 dev_kfree_skb_any(skb
);
390 tx_req
->mapping
= addr
;
392 if (unlikely(post_send(priv
, priv
->tx_head
& (ipoib_sendq_size
- 1),
393 address
->ah
, qpn
, addr
, skb
->len
))) {
394 ipoib_warn(priv
, "post_send failed\n");
395 ++priv
->stats
.tx_errors
;
396 ib_dma_unmap_single(priv
->ca
, addr
, skb
->len
, DMA_TO_DEVICE
);
397 dev_kfree_skb_any(skb
);
399 dev
->trans_start
= jiffies
;
401 address
->last_send
= priv
->tx_head
;
404 if (priv
->tx_head
- priv
->tx_tail
== ipoib_sendq_size
) {
405 ipoib_dbg(priv
, "TX ring full, stopping kernel net queue\n");
406 netif_stop_queue(dev
);
407 set_bit(IPOIB_FLAG_NETIF_STOPPED
, &priv
->flags
);
412 static void __ipoib_reap_ah(struct net_device
*dev
)
414 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
415 struct ipoib_ah
*ah
, *tah
;
416 LIST_HEAD(remove_list
);
418 spin_lock_irq(&priv
->tx_lock
);
419 spin_lock(&priv
->lock
);
420 list_for_each_entry_safe(ah
, tah
, &priv
->dead_ahs
, list
)
421 if ((int) priv
->tx_tail
- (int) ah
->last_send
>= 0) {
423 ib_destroy_ah(ah
->ah
);
426 spin_unlock(&priv
->lock
);
427 spin_unlock_irq(&priv
->tx_lock
);
430 void ipoib_reap_ah(struct work_struct
*work
)
432 struct ipoib_dev_priv
*priv
=
433 container_of(work
, struct ipoib_dev_priv
, ah_reap_task
.work
);
434 struct net_device
*dev
= priv
->dev
;
436 __ipoib_reap_ah(dev
);
438 if (!test_bit(IPOIB_STOP_REAPER
, &priv
->flags
))
439 queue_delayed_work(ipoib_workqueue
, &priv
->ah_reap_task
, HZ
);
442 int ipoib_ib_dev_open(struct net_device
*dev
)
444 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
447 if (ib_find_pkey(priv
->ca
, priv
->port
, priv
->pkey
, &priv
->pkey_index
)) {
448 ipoib_warn(priv
, "P_Key 0x%04x not found\n", priv
->pkey
);
449 clear_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
);
452 set_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
);
454 ret
= ipoib_init_qp(dev
);
456 ipoib_warn(priv
, "ipoib_init_qp returned %d\n", ret
);
460 ret
= ipoib_ib_post_receives(dev
);
462 ipoib_warn(priv
, "ipoib_ib_post_receives returned %d\n", ret
);
463 ipoib_ib_dev_stop(dev
, 1);
467 ret
= ipoib_cm_dev_open(dev
);
469 ipoib_warn(priv
, "ipoib_cm_dev_open returned %d\n", ret
);
470 ipoib_ib_dev_stop(dev
, 1);
474 clear_bit(IPOIB_STOP_REAPER
, &priv
->flags
);
475 queue_delayed_work(ipoib_workqueue
, &priv
->ah_reap_task
, HZ
);
477 set_bit(IPOIB_FLAG_INITIALIZED
, &priv
->flags
);
482 static void ipoib_pkey_dev_check_presence(struct net_device
*dev
)
484 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
487 if (ib_find_cached_pkey(priv
->ca
, priv
->port
, priv
->pkey
, &pkey_index
))
488 clear_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
);
490 set_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
);
493 int ipoib_ib_dev_up(struct net_device
*dev
)
495 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
497 ipoib_pkey_dev_check_presence(dev
);
499 if (!test_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
)) {
500 ipoib_dbg(priv
, "PKEY is not assigned.\n");
504 set_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
);
506 return ipoib_mcast_start_thread(dev
);
509 int ipoib_ib_dev_down(struct net_device
*dev
, int flush
)
511 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
513 ipoib_dbg(priv
, "downing ib_dev\n");
515 clear_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
);
516 netif_carrier_off(dev
);
518 /* Shutdown the P_Key thread if still active */
519 if (!test_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
)) {
520 mutex_lock(&pkey_mutex
);
521 set_bit(IPOIB_PKEY_STOP
, &priv
->flags
);
522 cancel_delayed_work(&priv
->pkey_poll_task
);
523 mutex_unlock(&pkey_mutex
);
525 flush_workqueue(ipoib_workqueue
);
528 ipoib_mcast_stop_thread(dev
, flush
);
529 ipoib_mcast_dev_flush(dev
);
531 ipoib_flush_paths(dev
);
536 static int recvs_pending(struct net_device
*dev
)
538 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
542 for (i
= 0; i
< ipoib_recvq_size
; ++i
)
543 if (priv
->rx_ring
[i
].skb
)
549 void ipoib_drain_cq(struct net_device
*dev
)
551 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
554 n
= ib_poll_cq(priv
->cq
, IPOIB_NUM_WC
, priv
->ibwc
);
555 for (i
= 0; i
< n
; ++i
) {
556 if (priv
->ibwc
[i
].wr_id
& IPOIB_CM_OP_SRQ
)
557 ipoib_cm_handle_rx_wc(dev
, priv
->ibwc
+ i
);
558 else if (priv
->ibwc
[i
].wr_id
& IPOIB_OP_RECV
)
559 ipoib_ib_handle_rx_wc(dev
, priv
->ibwc
+ i
);
561 ipoib_ib_handle_tx_wc(dev
, priv
->ibwc
+ i
);
563 } while (n
== IPOIB_NUM_WC
);
566 int ipoib_ib_dev_stop(struct net_device
*dev
, int flush
)
568 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
569 struct ib_qp_attr qp_attr
;
571 struct ipoib_tx_buf
*tx_req
;
574 clear_bit(IPOIB_FLAG_INITIALIZED
, &priv
->flags
);
576 ipoib_cm_dev_stop(dev
);
579 * Move our QP to the error state and then reinitialize in
580 * when all work requests have completed or have been flushed.
582 qp_attr
.qp_state
= IB_QPS_ERR
;
583 if (ib_modify_qp(priv
->qp
, &qp_attr
, IB_QP_STATE
))
584 ipoib_warn(priv
, "Failed to modify QP to ERROR state\n");
586 /* Wait for all sends and receives to complete */
589 while (priv
->tx_head
!= priv
->tx_tail
|| recvs_pending(dev
)) {
590 if (time_after(jiffies
, begin
+ 5 * HZ
)) {
591 ipoib_warn(priv
, "timing out; %d sends %d receives not completed\n",
592 priv
->tx_head
- priv
->tx_tail
, recvs_pending(dev
));
595 * assume the HW is wedged and just free up
596 * all our pending work requests.
598 while ((int) priv
->tx_tail
- (int) priv
->tx_head
< 0) {
599 tx_req
= &priv
->tx_ring
[priv
->tx_tail
&
600 (ipoib_sendq_size
- 1)];
601 ib_dma_unmap_single(priv
->ca
,
605 dev_kfree_skb_any(tx_req
->skb
);
609 for (i
= 0; i
< ipoib_recvq_size
; ++i
) {
610 struct ipoib_rx_buf
*rx_req
;
612 rx_req
= &priv
->rx_ring
[i
];
615 ib_dma_unmap_single(priv
->ca
,
619 dev_kfree_skb_any(rx_req
->skb
);
631 ipoib_dbg(priv
, "All sends and receives done.\n");
634 qp_attr
.qp_state
= IB_QPS_RESET
;
635 if (ib_modify_qp(priv
->qp
, &qp_attr
, IB_QP_STATE
))
636 ipoib_warn(priv
, "Failed to modify QP to RESET state\n");
638 /* Wait for all AHs to be reaped */
639 set_bit(IPOIB_STOP_REAPER
, &priv
->flags
);
640 cancel_delayed_work(&priv
->ah_reap_task
);
642 flush_workqueue(ipoib_workqueue
);
646 while (!list_empty(&priv
->dead_ahs
)) {
647 __ipoib_reap_ah(dev
);
649 if (time_after(jiffies
, begin
+ HZ
)) {
650 ipoib_warn(priv
, "timing out; will leak address handles\n");
657 ib_req_notify_cq(priv
->cq
, IB_CQ_NEXT_COMP
);
662 int ipoib_ib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
)
664 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
670 if (ipoib_transport_dev_init(dev
, ca
)) {
671 printk(KERN_WARNING
"%s: ipoib_transport_dev_init failed\n", ca
->name
);
675 if (dev
->flags
& IFF_UP
) {
676 if (ipoib_ib_dev_open(dev
)) {
677 ipoib_transport_dev_cleanup(dev
);
685 static void __ipoib_ib_dev_flush(struct ipoib_dev_priv
*priv
, int pkey_event
)
687 struct ipoib_dev_priv
*cpriv
;
688 struct net_device
*dev
= priv
->dev
;
691 mutex_lock(&priv
->vlan_mutex
);
694 * Flush any child interfaces too -- they might be up even if
695 * the parent is down.
697 list_for_each_entry(cpriv
, &priv
->child_intfs
, list
)
698 __ipoib_ib_dev_flush(cpriv
, pkey_event
);
700 mutex_unlock(&priv
->vlan_mutex
);
702 if (!test_bit(IPOIB_FLAG_INITIALIZED
, &priv
->flags
)) {
703 ipoib_dbg(priv
, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n");
707 if (!test_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
)) {
708 ipoib_dbg(priv
, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n");
713 if (ib_find_pkey(priv
->ca
, priv
->port
, priv
->pkey
, &new_index
)) {
714 clear_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
);
715 ipoib_ib_dev_down(dev
, 0);
716 ipoib_pkey_dev_delay_open(dev
);
719 set_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
);
721 /* restart QP only if P_Key index is changed */
722 if (new_index
== priv
->pkey_index
) {
723 ipoib_dbg(priv
, "Not flushing - P_Key index not changed.\n");
726 priv
->pkey_index
= new_index
;
729 ipoib_dbg(priv
, "flushing\n");
731 ipoib_ib_dev_down(dev
, 0);
734 ipoib_ib_dev_stop(dev
, 0);
735 ipoib_ib_dev_open(dev
);
739 * The device could have been brought down between the start and when
740 * we get here, don't bring it back up if it's not configured up
742 if (test_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
)) {
743 ipoib_ib_dev_up(dev
);
744 ipoib_mcast_restart_task(&priv
->restart_task
);
748 void ipoib_ib_dev_flush(struct work_struct
*work
)
750 struct ipoib_dev_priv
*priv
=
751 container_of(work
, struct ipoib_dev_priv
, flush_task
);
753 ipoib_dbg(priv
, "Flushing %s\n", priv
->dev
->name
);
754 __ipoib_ib_dev_flush(priv
, 0);
757 void ipoib_pkey_event(struct work_struct
*work
)
759 struct ipoib_dev_priv
*priv
=
760 container_of(work
, struct ipoib_dev_priv
, pkey_event_task
);
762 ipoib_dbg(priv
, "Flushing %s and restarting its QP\n", priv
->dev
->name
);
763 __ipoib_ib_dev_flush(priv
, 1);
766 void ipoib_ib_dev_cleanup(struct net_device
*dev
)
768 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
770 ipoib_dbg(priv
, "cleaning up ib_dev\n");
772 ipoib_mcast_stop_thread(dev
, 1);
773 ipoib_mcast_dev_flush(dev
);
775 ipoib_transport_dev_cleanup(dev
);
779 * Delayed P_Key Assigment Interim Support
781 * The following is initial implementation of delayed P_Key assigment
782 * mechanism. It is using the same approach implemented for the multicast
783 * group join. The single goal of this implementation is to quickly address
784 * Bug #2507. This implementation will probably be removed when the P_Key
785 * change async notification is available.
788 void ipoib_pkey_poll(struct work_struct
*work
)
790 struct ipoib_dev_priv
*priv
=
791 container_of(work
, struct ipoib_dev_priv
, pkey_poll_task
.work
);
792 struct net_device
*dev
= priv
->dev
;
794 ipoib_pkey_dev_check_presence(dev
);
796 if (test_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
))
799 mutex_lock(&pkey_mutex
);
800 if (!test_bit(IPOIB_PKEY_STOP
, &priv
->flags
))
801 queue_delayed_work(ipoib_workqueue
,
802 &priv
->pkey_poll_task
,
804 mutex_unlock(&pkey_mutex
);
808 int ipoib_pkey_dev_delay_open(struct net_device
*dev
)
810 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
812 /* Look for the interface pkey value in the IB Port P_Key table and */
813 /* set the interface pkey assigment flag */
814 ipoib_pkey_dev_check_presence(dev
);
816 /* P_Key value not assigned yet - start polling */
817 if (!test_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
)) {
818 mutex_lock(&pkey_mutex
);
819 clear_bit(IPOIB_PKEY_STOP
, &priv
->flags
);
820 queue_delayed_work(ipoib_workqueue
,
821 &priv
->pkey_poll_task
,
823 mutex_unlock(&pkey_mutex
);