]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. | |
2a1d9b7f RD |
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. | |
1da177e4 LT |
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 | ||
a4d61e84 | 41 | #include <rdma/ib_cache.h> |
1da177e4 LT |
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 | ||
95ed644f | 55 | static DEFINE_MUTEX(pkey_mutex); |
1da177e4 LT |
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 | ||
31c02e21 RD |
87 | spin_lock_irqsave(&priv->lock, flags); |
88 | list_add_tail(&ah->list, &priv->dead_ahs); | |
89 | spin_unlock_irqrestore(&priv->lock, flags); | |
1da177e4 LT |
90 | } |
91 | ||
1993d683 | 92 | static int ipoib_ib_post_receive(struct net_device *dev, int id) |
1da177e4 | 93 | { |
1993d683 RD |
94 | struct ipoib_dev_priv *priv = netdev_priv(dev); |
95 | struct ib_sge list; | |
96 | struct ib_recv_wr param; | |
1da177e4 | 97 | struct ib_recv_wr *bad_wr; |
1993d683 RD |
98 | int ret; |
99 | ||
100 | list.addr = priv->rx_ring[id].mapping; | |
101 | list.length = IPOIB_BUF_SIZE; | |
102 | list.lkey = priv->mr->lkey; | |
103 | ||
104 | param.next = NULL; | |
105 | param.wr_id = id | IPOIB_OP_RECV; | |
106 | param.sg_list = &list; | |
107 | param.num_sge = 1; | |
108 | ||
109 | ret = ib_post_recv(priv->qp, ¶m, &bad_wr); | |
110 | if (unlikely(ret)) { | |
111 | ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret); | |
112 | dma_unmap_single(priv->ca->dma_device, | |
113 | priv->rx_ring[id].mapping, | |
114 | IPOIB_BUF_SIZE, DMA_FROM_DEVICE); | |
115 | dev_kfree_skb_any(priv->rx_ring[id].skb); | |
116 | priv->rx_ring[id].skb = NULL; | |
117 | } | |
1da177e4 | 118 | |
1993d683 | 119 | return ret; |
1da177e4 LT |
120 | } |
121 | ||
1993d683 | 122 | static int ipoib_alloc_rx_skb(struct net_device *dev, int id) |
1da177e4 LT |
123 | { |
124 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
125 | struct sk_buff *skb; | |
126 | dma_addr_t addr; | |
1da177e4 LT |
127 | |
128 | skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4); | |
1993d683 | 129 | if (!skb) |
1da177e4 | 130 | return -ENOMEM; |
1993d683 RD |
131 | |
132 | /* | |
133 | * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte | |
134 | * header. So we need 4 more bytes to get to 48 and align the | |
135 | * IP header to a multiple of 16. | |
136 | */ | |
137 | skb_reserve(skb, 4); | |
138 | ||
1da177e4 LT |
139 | addr = dma_map_single(priv->ca->dma_device, |
140 | skb->data, IPOIB_BUF_SIZE, | |
141 | DMA_FROM_DEVICE); | |
1993d683 | 142 | if (unlikely(dma_mapping_error(addr))) { |
1da177e4 | 143 | dev_kfree_skb_any(skb); |
1993d683 | 144 | return -EIO; |
1da177e4 LT |
145 | } |
146 | ||
1993d683 RD |
147 | priv->rx_ring[id].skb = skb; |
148 | priv->rx_ring[id].mapping = addr; | |
149 | ||
150 | return 0; | |
1da177e4 LT |
151 | } |
152 | ||
153 | static int ipoib_ib_post_receives(struct net_device *dev) | |
154 | { | |
155 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
156 | int i; | |
157 | ||
0f485251 | 158 | for (i = 0; i < ipoib_recvq_size; ++i) { |
1993d683 RD |
159 | if (ipoib_alloc_rx_skb(dev, i)) { |
160 | ipoib_warn(priv, "failed to allocate receive buffer %d\n", i); | |
161 | return -ENOMEM; | |
162 | } | |
1da177e4 LT |
163 | if (ipoib_ib_post_receive(dev, i)) { |
164 | ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i); | |
165 | return -EIO; | |
166 | } | |
167 | } | |
168 | ||
169 | return 0; | |
170 | } | |
171 | ||
2439a6e6 | 172 | static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) |
1da177e4 LT |
173 | { |
174 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
2439a6e6 RD |
175 | unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV; |
176 | struct sk_buff *skb; | |
177 | dma_addr_t addr; | |
1da177e4 | 178 | |
2439a6e6 | 179 | ipoib_dbg_data(priv, "recv completion: id %d, op %d, status: %d\n", |
1da177e4 LT |
180 | wr_id, wc->opcode, wc->status); |
181 | ||
2439a6e6 RD |
182 | if (unlikely(wr_id >= ipoib_recvq_size)) { |
183 | ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n", | |
184 | wr_id, ipoib_recvq_size); | |
185 | return; | |
186 | } | |
187 | ||
188 | skb = priv->rx_ring[wr_id].skb; | |
189 | addr = priv->rx_ring[wr_id].mapping; | |
190 | ||
191 | if (unlikely(wc->status != IB_WC_SUCCESS)) { | |
192 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
193 | ipoib_warn(priv, "failed recv event " | |
194 | "(status=%d, wrid=%d vend_err %x)\n", | |
195 | wc->status, wr_id, wc->vendor_err); | |
196 | dma_unmap_single(priv->ca->dma_device, addr, | |
197 | IPOIB_BUF_SIZE, DMA_FROM_DEVICE); | |
198 | dev_kfree_skb_any(skb); | |
199 | priv->rx_ring[wr_id].skb = NULL; | |
200 | return; | |
201 | } | |
1da177e4 | 202 | |
2439a6e6 RD |
203 | /* |
204 | * If we can't allocate a new RX buffer, dump | |
205 | * this packet and reuse the old buffer. | |
206 | */ | |
207 | if (unlikely(ipoib_alloc_rx_skb(dev, wr_id))) { | |
208 | ++priv->stats.rx_dropped; | |
209 | goto repost; | |
210 | } | |
211 | ||
212 | ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n", | |
213 | wc->byte_len, wc->slid); | |
214 | ||
215 | dma_unmap_single(priv->ca->dma_device, addr, | |
216 | IPOIB_BUF_SIZE, DMA_FROM_DEVICE); | |
217 | ||
218 | skb_put(skb, wc->byte_len); | |
219 | skb_pull(skb, IB_GRH_BYTES); | |
220 | ||
221 | if (wc->slid != priv->local_lid || | |
222 | wc->src_qp != priv->qp->qp_num) { | |
223 | skb->protocol = ((struct ipoib_header *) skb->data)->proto; | |
224 | skb->mac.raw = skb->data; | |
225 | skb_pull(skb, IPOIB_ENCAP_LEN); | |
226 | ||
227 | dev->last_rx = jiffies; | |
228 | ++priv->stats.rx_packets; | |
229 | priv->stats.rx_bytes += skb->len; | |
230 | ||
231 | skb->dev = dev; | |
232 | /* XXX get correct PACKET_ type here */ | |
233 | skb->pkt_type = PACKET_HOST; | |
234 | netif_rx_ni(skb); | |
1da177e4 | 235 | } else { |
2439a6e6 RD |
236 | ipoib_dbg_data(priv, "dropping loopback packet\n"); |
237 | dev_kfree_skb_any(skb); | |
238 | } | |
1da177e4 | 239 | |
2439a6e6 RD |
240 | repost: |
241 | if (unlikely(ipoib_ib_post_receive(dev, wr_id))) | |
242 | ipoib_warn(priv, "ipoib_ib_post_receive failed " | |
243 | "for buf %d\n", wr_id); | |
244 | } | |
1da177e4 | 245 | |
2439a6e6 RD |
246 | static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) |
247 | { | |
248 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
249 | unsigned int wr_id = wc->wr_id; | |
250 | struct ipoib_tx_buf *tx_req; | |
251 | unsigned long flags; | |
1da177e4 | 252 | |
2439a6e6 RD |
253 | ipoib_dbg_data(priv, "send completion: id %d, op %d, status: %d\n", |
254 | wr_id, wc->opcode, wc->status); | |
1da177e4 | 255 | |
2439a6e6 RD |
256 | if (unlikely(wr_id >= ipoib_sendq_size)) { |
257 | ipoib_warn(priv, "send completion event with wrid %d (> %d)\n", | |
258 | wr_id, ipoib_sendq_size); | |
259 | return; | |
1da177e4 | 260 | } |
2439a6e6 RD |
261 | |
262 | tx_req = &priv->tx_ring[wr_id]; | |
263 | ||
264 | dma_unmap_single(priv->ca->dma_device, | |
265 | pci_unmap_addr(tx_req, mapping), | |
266 | tx_req->skb->len, | |
267 | DMA_TO_DEVICE); | |
268 | ||
269 | ++priv->stats.tx_packets; | |
270 | priv->stats.tx_bytes += tx_req->skb->len; | |
271 | ||
272 | dev_kfree_skb_any(tx_req->skb); | |
273 | ||
274 | spin_lock_irqsave(&priv->tx_lock, flags); | |
275 | ++priv->tx_tail; | |
276 | if (netif_queue_stopped(dev) && | |
277 | test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags) && | |
278 | priv->tx_head - priv->tx_tail <= ipoib_sendq_size >> 1) | |
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 | static void ipoib_ib_handle_wc(struct net_device *dev, struct ib_wc *wc) | |
290 | { | |
291 | if (wc->wr_id & IPOIB_OP_RECV) | |
292 | ipoib_ib_handle_rx_wc(dev, wc); | |
293 | else | |
294 | ipoib_ib_handle_tx_wc(dev, wc); | |
1da177e4 LT |
295 | } |
296 | ||
297 | void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr) | |
298 | { | |
299 | struct net_device *dev = (struct net_device *) dev_ptr; | |
300 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
301 | int n, i; | |
302 | ||
303 | ib_req_notify_cq(cq, IB_CQ_NEXT_COMP); | |
304 | do { | |
305 | n = ib_poll_cq(cq, IPOIB_NUM_WC, priv->ibwc); | |
306 | for (i = 0; i < n; ++i) | |
307 | ipoib_ib_handle_wc(dev, priv->ibwc + i); | |
308 | } while (n == IPOIB_NUM_WC); | |
309 | } | |
310 | ||
311 | static inline int post_send(struct ipoib_dev_priv *priv, | |
312 | unsigned int wr_id, | |
313 | struct ib_ah *address, u32 qpn, | |
314 | dma_addr_t addr, int len) | |
315 | { | |
316 | struct ib_send_wr *bad_wr; | |
317 | ||
318 | priv->tx_sge.addr = addr; | |
319 | priv->tx_sge.length = len; | |
320 | ||
321 | priv->tx_wr.wr_id = wr_id; | |
322 | priv->tx_wr.wr.ud.remote_qpn = qpn; | |
323 | priv->tx_wr.wr.ud.ah = address; | |
324 | ||
325 | return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr); | |
326 | } | |
327 | ||
328 | void ipoib_send(struct net_device *dev, struct sk_buff *skb, | |
329 | struct ipoib_ah *address, u32 qpn) | |
330 | { | |
331 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1993d683 | 332 | struct ipoib_tx_buf *tx_req; |
1da177e4 LT |
333 | dma_addr_t addr; |
334 | ||
a8bfca02 | 335 | if (unlikely(skb->len > dev->mtu + INFINIBAND_ALEN)) { |
1da177e4 LT |
336 | ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n", |
337 | skb->len, dev->mtu + INFINIBAND_ALEN); | |
338 | ++priv->stats.tx_dropped; | |
339 | ++priv->stats.tx_errors; | |
340 | dev_kfree_skb_any(skb); | |
341 | return; | |
342 | } | |
343 | ||
344 | ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n", | |
345 | skb->len, address, qpn); | |
346 | ||
347 | /* | |
348 | * We put the skb into the tx_ring _before_ we call post_send() | |
349 | * because it's entirely possible that the completion handler will | |
350 | * run before we execute anything after the post_send(). That | |
351 | * means we have to make sure everything is properly recorded and | |
352 | * our state is consistent before we call post_send(). | |
353 | */ | |
0f485251 | 354 | tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)]; |
1da177e4 LT |
355 | tx_req->skb = skb; |
356 | addr = dma_map_single(priv->ca->dma_device, skb->data, skb->len, | |
357 | DMA_TO_DEVICE); | |
73fbe8be RD |
358 | if (unlikely(dma_mapping_error(addr))) { |
359 | ++priv->stats.tx_errors; | |
360 | dev_kfree_skb_any(skb); | |
361 | return; | |
362 | } | |
1da177e4 LT |
363 | pci_unmap_addr_set(tx_req, mapping, addr); |
364 | ||
0f485251 | 365 | if (unlikely(post_send(priv, priv->tx_head & (ipoib_sendq_size - 1), |
1da177e4 LT |
366 | address->ah, qpn, addr, skb->len))) { |
367 | ipoib_warn(priv, "post_send failed\n"); | |
368 | ++priv->stats.tx_errors; | |
369 | dma_unmap_single(priv->ca->dma_device, addr, skb->len, | |
370 | DMA_TO_DEVICE); | |
371 | dev_kfree_skb_any(skb); | |
372 | } else { | |
373 | dev->trans_start = jiffies; | |
374 | ||
375 | address->last_send = priv->tx_head; | |
376 | ++priv->tx_head; | |
377 | ||
0f485251 | 378 | if (priv->tx_head - priv->tx_tail == ipoib_sendq_size) { |
1da177e4 LT |
379 | ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n"); |
380 | netif_stop_queue(dev); | |
381 | } | |
382 | } | |
383 | } | |
384 | ||
385 | static void __ipoib_reap_ah(struct net_device *dev) | |
386 | { | |
387 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
388 | struct ipoib_ah *ah, *tah; | |
389 | LIST_HEAD(remove_list); | |
390 | ||
31c02e21 RD |
391 | spin_lock_irq(&priv->tx_lock); |
392 | spin_lock(&priv->lock); | |
1da177e4 | 393 | list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list) |
2181858b | 394 | if ((int) priv->tx_tail - (int) ah->last_send >= 0) { |
1da177e4 | 395 | list_del(&ah->list); |
31c02e21 RD |
396 | ib_destroy_ah(ah->ah); |
397 | kfree(ah); | |
1da177e4 | 398 | } |
31c02e21 RD |
399 | spin_unlock(&priv->lock); |
400 | spin_unlock_irq(&priv->tx_lock); | |
1da177e4 LT |
401 | } |
402 | ||
c4028958 | 403 | void ipoib_reap_ah(struct work_struct *work) |
1da177e4 | 404 | { |
c4028958 DH |
405 | struct ipoib_dev_priv *priv = |
406 | container_of(work, struct ipoib_dev_priv, ah_reap_task.work); | |
407 | struct net_device *dev = priv->dev; | |
1da177e4 LT |
408 | |
409 | __ipoib_reap_ah(dev); | |
410 | ||
411 | if (!test_bit(IPOIB_STOP_REAPER, &priv->flags)) | |
412 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ); | |
413 | } | |
414 | ||
415 | int ipoib_ib_dev_open(struct net_device *dev) | |
416 | { | |
417 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
418 | int ret; | |
419 | ||
5b6810e0 | 420 | ret = ipoib_init_qp(dev); |
1da177e4 | 421 | if (ret) { |
5b6810e0 | 422 | ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret); |
1da177e4 LT |
423 | return -1; |
424 | } | |
425 | ||
426 | ret = ipoib_ib_post_receives(dev); | |
427 | if (ret) { | |
428 | ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret); | |
54d07e2a | 429 | ipoib_ib_dev_stop(dev); |
1da177e4 LT |
430 | return -1; |
431 | } | |
432 | ||
433 | clear_bit(IPOIB_STOP_REAPER, &priv->flags); | |
434 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ); | |
435 | ||
7a343d4c LA |
436 | set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags); |
437 | ||
1da177e4 LT |
438 | return 0; |
439 | } | |
440 | ||
7a343d4c LA |
441 | static void ipoib_pkey_dev_check_presence(struct net_device *dev) |
442 | { | |
443 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
444 | u16 pkey_index = 0; | |
445 | ||
446 | if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) | |
447 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
448 | else | |
449 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
450 | } | |
451 | ||
1da177e4 LT |
452 | int ipoib_ib_dev_up(struct net_device *dev) |
453 | { | |
454 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
455 | ||
7a343d4c LA |
456 | ipoib_pkey_dev_check_presence(dev); |
457 | ||
458 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
459 | ipoib_dbg(priv, "PKEY is not assigned.\n"); | |
460 | return 0; | |
461 | } | |
462 | ||
1da177e4 LT |
463 | set_bit(IPOIB_FLAG_OPER_UP, &priv->flags); |
464 | ||
465 | return ipoib_mcast_start_thread(dev); | |
466 | } | |
467 | ||
0b3ea082 | 468 | int ipoib_ib_dev_down(struct net_device *dev, int flush) |
1da177e4 LT |
469 | { |
470 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
471 | ||
472 | ipoib_dbg(priv, "downing ib_dev\n"); | |
473 | ||
474 | clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags); | |
475 | netif_carrier_off(dev); | |
476 | ||
477 | /* Shutdown the P_Key thread if still active */ | |
478 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
95ed644f | 479 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
480 | set_bit(IPOIB_PKEY_STOP, &priv->flags); |
481 | cancel_delayed_work(&priv->pkey_task); | |
95ed644f | 482 | mutex_unlock(&pkey_mutex); |
0b3ea082 JM |
483 | if (flush) |
484 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
485 | } |
486 | ||
0b3ea082 | 487 | ipoib_mcast_stop_thread(dev, flush); |
1da177e4 LT |
488 | ipoib_mcast_dev_flush(dev); |
489 | ||
1da177e4 LT |
490 | ipoib_flush_paths(dev); |
491 | ||
492 | return 0; | |
493 | } | |
494 | ||
495 | static int recvs_pending(struct net_device *dev) | |
496 | { | |
497 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
498 | int pending = 0; | |
499 | int i; | |
500 | ||
0f485251 | 501 | for (i = 0; i < ipoib_recvq_size; ++i) |
1da177e4 LT |
502 | if (priv->rx_ring[i].skb) |
503 | ++pending; | |
504 | ||
505 | return pending; | |
506 | } | |
507 | ||
508 | int ipoib_ib_dev_stop(struct net_device *dev) | |
509 | { | |
510 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
511 | struct ib_qp_attr qp_attr; | |
1da177e4 | 512 | unsigned long begin; |
1993d683 | 513 | struct ipoib_tx_buf *tx_req; |
1da177e4 LT |
514 | int i; |
515 | ||
7a343d4c LA |
516 | clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags); |
517 | ||
3bc12e75 RD |
518 | /* |
519 | * Move our QP to the error state and then reinitialize in | |
520 | * when all work requests have completed or have been flushed. | |
521 | */ | |
1da177e4 | 522 | qp_attr.qp_state = IB_QPS_ERR; |
3bc12e75 | 523 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
524 | ipoib_warn(priv, "Failed to modify QP to ERROR state\n"); |
525 | ||
526 | /* Wait for all sends and receives to complete */ | |
527 | begin = jiffies; | |
528 | ||
529 | while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) { | |
530 | if (time_after(jiffies, begin + 5 * HZ)) { | |
531 | ipoib_warn(priv, "timing out; %d sends %d receives not completed\n", | |
532 | priv->tx_head - priv->tx_tail, recvs_pending(dev)); | |
533 | ||
534 | /* | |
535 | * assume the HW is wedged and just free up | |
536 | * all our pending work requests. | |
537 | */ | |
2181858b | 538 | while ((int) priv->tx_tail - (int) priv->tx_head < 0) { |
1da177e4 | 539 | tx_req = &priv->tx_ring[priv->tx_tail & |
0f485251 | 540 | (ipoib_sendq_size - 1)]; |
1da177e4 LT |
541 | dma_unmap_single(priv->ca->dma_device, |
542 | pci_unmap_addr(tx_req, mapping), | |
543 | tx_req->skb->len, | |
544 | DMA_TO_DEVICE); | |
545 | dev_kfree_skb_any(tx_req->skb); | |
546 | ++priv->tx_tail; | |
547 | } | |
548 | ||
0f485251 | 549 | for (i = 0; i < ipoib_recvq_size; ++i) |
1da177e4 LT |
550 | if (priv->rx_ring[i].skb) { |
551 | dma_unmap_single(priv->ca->dma_device, | |
552 | pci_unmap_addr(&priv->rx_ring[i], | |
553 | mapping), | |
554 | IPOIB_BUF_SIZE, | |
555 | DMA_FROM_DEVICE); | |
556 | dev_kfree_skb_any(priv->rx_ring[i].skb); | |
557 | priv->rx_ring[i].skb = NULL; | |
558 | } | |
559 | ||
560 | goto timeout; | |
561 | } | |
562 | ||
563 | msleep(1); | |
564 | } | |
565 | ||
566 | ipoib_dbg(priv, "All sends and receives done.\n"); | |
567 | ||
568 | timeout: | |
569 | qp_attr.qp_state = IB_QPS_RESET; | |
3bc12e75 | 570 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
571 | ipoib_warn(priv, "Failed to modify QP to RESET state\n"); |
572 | ||
573 | /* Wait for all AHs to be reaped */ | |
574 | set_bit(IPOIB_STOP_REAPER, &priv->flags); | |
575 | cancel_delayed_work(&priv->ah_reap_task); | |
576 | flush_workqueue(ipoib_workqueue); | |
577 | ||
578 | begin = jiffies; | |
579 | ||
580 | while (!list_empty(&priv->dead_ahs)) { | |
581 | __ipoib_reap_ah(dev); | |
582 | ||
583 | if (time_after(jiffies, begin + HZ)) { | |
584 | ipoib_warn(priv, "timing out; will leak address handles\n"); | |
585 | break; | |
586 | } | |
587 | ||
588 | msleep(1); | |
589 | } | |
590 | ||
591 | return 0; | |
592 | } | |
593 | ||
594 | int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port) | |
595 | { | |
596 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
597 | ||
598 | priv->ca = ca; | |
599 | priv->port = port; | |
600 | priv->qp = NULL; | |
601 | ||
602 | if (ipoib_transport_dev_init(dev, ca)) { | |
603 | printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name); | |
604 | return -ENODEV; | |
605 | } | |
606 | ||
607 | if (dev->flags & IFF_UP) { | |
608 | if (ipoib_ib_dev_open(dev)) { | |
609 | ipoib_transport_dev_cleanup(dev); | |
610 | return -ENODEV; | |
611 | } | |
612 | } | |
613 | ||
614 | return 0; | |
615 | } | |
616 | ||
c4028958 | 617 | void ipoib_ib_dev_flush(struct work_struct *work) |
1da177e4 | 618 | { |
c4028958 DH |
619 | struct ipoib_dev_priv *cpriv, *priv = |
620 | container_of(work, struct ipoib_dev_priv, flush_task); | |
621 | struct net_device *dev = priv->dev; | |
1da177e4 | 622 | |
7a343d4c LA |
623 | if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags) ) { |
624 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n"); | |
625 | return; | |
626 | } | |
627 | ||
628 | if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { | |
629 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n"); | |
1da177e4 | 630 | return; |
7a343d4c | 631 | } |
1da177e4 LT |
632 | |
633 | ipoib_dbg(priv, "flushing\n"); | |
634 | ||
0b3ea082 | 635 | ipoib_ib_dev_down(dev, 0); |
1da177e4 LT |
636 | |
637 | /* | |
638 | * The device could have been brought down between the start and when | |
639 | * we get here, don't bring it back up if it's not configured up | |
640 | */ | |
5ccd0255 | 641 | if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { |
1da177e4 | 642 | ipoib_ib_dev_up(dev); |
c4028958 | 643 | ipoib_mcast_restart_task(&priv->restart_task); |
5ccd0255 | 644 | } |
1da177e4 | 645 | |
95ed644f | 646 | mutex_lock(&priv->vlan_mutex); |
4f71055a | 647 | |
1da177e4 LT |
648 | /* Flush any child interfaces too */ |
649 | list_for_each_entry(cpriv, &priv->child_intfs, list) | |
c4028958 | 650 | ipoib_ib_dev_flush(&cpriv->flush_task); |
4f71055a | 651 | |
95ed644f | 652 | mutex_unlock(&priv->vlan_mutex); |
1da177e4 LT |
653 | } |
654 | ||
655 | void ipoib_ib_dev_cleanup(struct net_device *dev) | |
656 | { | |
657 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
658 | ||
659 | ipoib_dbg(priv, "cleaning up ib_dev\n"); | |
660 | ||
8d2cae06 | 661 | ipoib_mcast_stop_thread(dev, 1); |
988bd503 | 662 | ipoib_mcast_dev_flush(dev); |
1da177e4 LT |
663 | |
664 | ipoib_transport_dev_cleanup(dev); | |
665 | } | |
666 | ||
667 | /* | |
668 | * Delayed P_Key Assigment Interim Support | |
669 | * | |
670 | * The following is initial implementation of delayed P_Key assigment | |
671 | * mechanism. It is using the same approach implemented for the multicast | |
672 | * group join. The single goal of this implementation is to quickly address | |
673 | * Bug #2507. This implementation will probably be removed when the P_Key | |
674 | * change async notification is available. | |
675 | */ | |
1da177e4 | 676 | |
c4028958 | 677 | void ipoib_pkey_poll(struct work_struct *work) |
1da177e4 | 678 | { |
c4028958 DH |
679 | struct ipoib_dev_priv *priv = |
680 | container_of(work, struct ipoib_dev_priv, pkey_task.work); | |
681 | struct net_device *dev = priv->dev; | |
1da177e4 LT |
682 | |
683 | ipoib_pkey_dev_check_presence(dev); | |
684 | ||
685 | if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) | |
686 | ipoib_open(dev); | |
687 | else { | |
95ed644f | 688 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
689 | if (!test_bit(IPOIB_PKEY_STOP, &priv->flags)) |
690 | queue_delayed_work(ipoib_workqueue, | |
691 | &priv->pkey_task, | |
692 | HZ); | |
95ed644f | 693 | mutex_unlock(&pkey_mutex); |
1da177e4 LT |
694 | } |
695 | } | |
696 | ||
697 | int ipoib_pkey_dev_delay_open(struct net_device *dev) | |
698 | { | |
699 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
700 | ||
701 | /* Look for the interface pkey value in the IB Port P_Key table and */ | |
702 | /* set the interface pkey assigment flag */ | |
703 | ipoib_pkey_dev_check_presence(dev); | |
704 | ||
705 | /* P_Key value not assigned yet - start polling */ | |
706 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
95ed644f | 707 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
708 | clear_bit(IPOIB_PKEY_STOP, &priv->flags); |
709 | queue_delayed_work(ipoib_workqueue, | |
710 | &priv->pkey_task, | |
711 | HZ); | |
95ed644f | 712 | mutex_unlock(&pkey_mutex); |
1da177e4 LT |
713 | return 1; |
714 | } | |
715 | ||
716 | return 0; | |
717 | } |