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. | |
1da177e4 LT |
34 | */ |
35 | ||
36 | #include <linux/delay.h> | |
37 | #include <linux/dma-mapping.h> | |
5a0e3ad6 | 38 | #include <linux/slab.h> |
1da177e4 | 39 | |
40ca1988 EC |
40 | #include <linux/ip.h> |
41 | #include <linux/tcp.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 | ||
95ed644f | 53 | static DEFINE_MUTEX(pkey_mutex); |
1da177e4 LT |
54 | |
55 | struct ipoib_ah *ipoib_create_ah(struct net_device *dev, | |
56 | struct ib_pd *pd, struct ib_ah_attr *attr) | |
57 | { | |
58 | struct ipoib_ah *ah; | |
59 | ||
60 | ah = kmalloc(sizeof *ah, GFP_KERNEL); | |
61 | if (!ah) | |
62 | return NULL; | |
63 | ||
64 | ah->dev = dev; | |
65 | ah->last_send = 0; | |
66 | kref_init(&ah->ref); | |
67 | ||
68 | ah->ah = ib_create_ah(pd, attr); | |
69 | if (IS_ERR(ah->ah)) { | |
70 | kfree(ah); | |
71 | ah = NULL; | |
72 | } else | |
73 | ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah); | |
74 | ||
75 | return ah; | |
76 | } | |
77 | ||
78 | void ipoib_free_ah(struct kref *kref) | |
79 | { | |
80 | struct ipoib_ah *ah = container_of(kref, struct ipoib_ah, ref); | |
81 | struct ipoib_dev_priv *priv = netdev_priv(ah->dev); | |
82 | ||
83 | unsigned long flags; | |
84 | ||
31c02e21 RD |
85 | spin_lock_irqsave(&priv->lock, flags); |
86 | list_add_tail(&ah->list, &priv->dead_ahs); | |
87 | spin_unlock_irqrestore(&priv->lock, flags); | |
1da177e4 LT |
88 | } |
89 | ||
bc7b3a36 SM |
90 | static void ipoib_ud_dma_unmap_rx(struct ipoib_dev_priv *priv, |
91 | u64 mapping[IPOIB_UD_RX_SG]) | |
92 | { | |
93 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) { | |
94 | ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_UD_HEAD_SIZE, | |
95 | DMA_FROM_DEVICE); | |
96 | ib_dma_unmap_page(priv->ca, mapping[1], PAGE_SIZE, | |
97 | DMA_FROM_DEVICE); | |
98 | } else | |
99 | ib_dma_unmap_single(priv->ca, mapping[0], | |
100 | IPOIB_UD_BUF_SIZE(priv->max_ib_mtu), | |
101 | DMA_FROM_DEVICE); | |
102 | } | |
103 | ||
104 | static void ipoib_ud_skb_put_frags(struct ipoib_dev_priv *priv, | |
105 | struct sk_buff *skb, | |
106 | unsigned int length) | |
107 | { | |
108 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) { | |
109 | skb_frag_t *frag = &skb_shinfo(skb)->frags[0]; | |
110 | unsigned int size; | |
111 | /* | |
112 | * There is only two buffers needed for max_payload = 4K, | |
113 | * first buf size is IPOIB_UD_HEAD_SIZE | |
114 | */ | |
115 | skb->tail += IPOIB_UD_HEAD_SIZE; | |
116 | skb->len += length; | |
117 | ||
118 | size = length - IPOIB_UD_HEAD_SIZE; | |
119 | ||
120 | frag->size = size; | |
121 | skb->data_len += size; | |
122 | skb->truesize += size; | |
123 | } else | |
124 | skb_put(skb, length); | |
125 | ||
126 | } | |
127 | ||
1993d683 | 128 | static int ipoib_ib_post_receive(struct net_device *dev, int id) |
1da177e4 | 129 | { |
1993d683 | 130 | struct ipoib_dev_priv *priv = netdev_priv(dev); |
1da177e4 | 131 | struct ib_recv_wr *bad_wr; |
1993d683 RD |
132 | int ret; |
133 | ||
bc7b3a36 SM |
134 | priv->rx_wr.wr_id = id | IPOIB_OP_RECV; |
135 | priv->rx_sge[0].addr = priv->rx_ring[id].mapping[0]; | |
136 | priv->rx_sge[1].addr = priv->rx_ring[id].mapping[1]; | |
1993d683 | 137 | |
1993d683 | 138 | |
bc7b3a36 | 139 | ret = ib_post_recv(priv->qp, &priv->rx_wr, &bad_wr); |
1993d683 RD |
140 | if (unlikely(ret)) { |
141 | ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret); | |
bc7b3a36 | 142 | ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[id].mapping); |
1993d683 RD |
143 | dev_kfree_skb_any(priv->rx_ring[id].skb); |
144 | priv->rx_ring[id].skb = NULL; | |
145 | } | |
1da177e4 | 146 | |
1993d683 | 147 | return ret; |
1da177e4 LT |
148 | } |
149 | ||
bc7b3a36 | 150 | static struct sk_buff *ipoib_alloc_rx_skb(struct net_device *dev, int id) |
1da177e4 LT |
151 | { |
152 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
153 | struct sk_buff *skb; | |
bc7b3a36 SM |
154 | int buf_size; |
155 | u64 *mapping; | |
1da177e4 | 156 | |
bc7b3a36 SM |
157 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) |
158 | buf_size = IPOIB_UD_HEAD_SIZE; | |
159 | else | |
160 | buf_size = IPOIB_UD_BUF_SIZE(priv->max_ib_mtu); | |
161 | ||
162 | skb = dev_alloc_skb(buf_size + 4); | |
163 | if (unlikely(!skb)) | |
164 | return NULL; | |
1993d683 RD |
165 | |
166 | /* | |
167 | * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte | |
168 | * header. So we need 4 more bytes to get to 48 and align the | |
169 | * IP header to a multiple of 16. | |
170 | */ | |
171 | skb_reserve(skb, 4); | |
172 | ||
bc7b3a36 SM |
173 | mapping = priv->rx_ring[id].mapping; |
174 | mapping[0] = ib_dma_map_single(priv->ca, skb->data, buf_size, | |
175 | DMA_FROM_DEVICE); | |
176 | if (unlikely(ib_dma_mapping_error(priv->ca, mapping[0]))) | |
177 | goto error; | |
178 | ||
179 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) { | |
180 | struct page *page = alloc_page(GFP_ATOMIC); | |
181 | if (!page) | |
182 | goto partial_error; | |
183 | skb_fill_page_desc(skb, 0, page, 0, PAGE_SIZE); | |
184 | mapping[1] = | |
185 | ib_dma_map_page(priv->ca, skb_shinfo(skb)->frags[0].page, | |
186 | 0, PAGE_SIZE, DMA_FROM_DEVICE); | |
187 | if (unlikely(ib_dma_mapping_error(priv->ca, mapping[1]))) | |
188 | goto partial_error; | |
1da177e4 LT |
189 | } |
190 | ||
bc7b3a36 SM |
191 | priv->rx_ring[id].skb = skb; |
192 | return skb; | |
1993d683 | 193 | |
bc7b3a36 SM |
194 | partial_error: |
195 | ib_dma_unmap_single(priv->ca, mapping[0], buf_size, DMA_FROM_DEVICE); | |
196 | error: | |
197 | dev_kfree_skb_any(skb); | |
198 | return NULL; | |
1da177e4 LT |
199 | } |
200 | ||
201 | static int ipoib_ib_post_receives(struct net_device *dev) | |
202 | { | |
203 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
204 | int i; | |
205 | ||
0f485251 | 206 | for (i = 0; i < ipoib_recvq_size; ++i) { |
bc7b3a36 | 207 | if (!ipoib_alloc_rx_skb(dev, i)) { |
1993d683 RD |
208 | ipoib_warn(priv, "failed to allocate receive buffer %d\n", i); |
209 | return -ENOMEM; | |
210 | } | |
1da177e4 LT |
211 | if (ipoib_ib_post_receive(dev, i)) { |
212 | ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i); | |
213 | return -EIO; | |
214 | } | |
215 | } | |
216 | ||
217 | return 0; | |
218 | } | |
219 | ||
2439a6e6 | 220 | static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) |
1da177e4 LT |
221 | { |
222 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
2439a6e6 RD |
223 | unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV; |
224 | struct sk_buff *skb; | |
bc7b3a36 | 225 | u64 mapping[IPOIB_UD_RX_SG]; |
1da177e4 | 226 | |
a89875fc RD |
227 | ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n", |
228 | wr_id, wc->status); | |
1da177e4 | 229 | |
2439a6e6 RD |
230 | if (unlikely(wr_id >= ipoib_recvq_size)) { |
231 | ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n", | |
232 | wr_id, ipoib_recvq_size); | |
233 | return; | |
234 | } | |
235 | ||
236 | skb = priv->rx_ring[wr_id].skb; | |
2439a6e6 RD |
237 | |
238 | if (unlikely(wc->status != IB_WC_SUCCESS)) { | |
239 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
240 | ipoib_warn(priv, "failed recv event " | |
241 | "(status=%d, wrid=%d vend_err %x)\n", | |
242 | wc->status, wr_id, wc->vendor_err); | |
bc7b3a36 | 243 | ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[wr_id].mapping); |
2439a6e6 RD |
244 | dev_kfree_skb_any(skb); |
245 | priv->rx_ring[wr_id].skb = NULL; | |
246 | return; | |
247 | } | |
1da177e4 | 248 | |
1b844afe RD |
249 | /* |
250 | * Drop packets that this interface sent, ie multicast packets | |
251 | * that the HCA has replicated. | |
252 | */ | |
253 | if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num) | |
254 | goto repost; | |
255 | ||
bc7b3a36 SM |
256 | memcpy(mapping, priv->rx_ring[wr_id].mapping, |
257 | IPOIB_UD_RX_SG * sizeof *mapping); | |
258 | ||
2439a6e6 RD |
259 | /* |
260 | * If we can't allocate a new RX buffer, dump | |
261 | * this packet and reuse the old buffer. | |
262 | */ | |
bc7b3a36 | 263 | if (unlikely(!ipoib_alloc_rx_skb(dev, wr_id))) { |
de903512 | 264 | ++dev->stats.rx_dropped; |
2439a6e6 RD |
265 | goto repost; |
266 | } | |
267 | ||
268 | ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n", | |
269 | wc->byte_len, wc->slid); | |
270 | ||
bc7b3a36 SM |
271 | ipoib_ud_dma_unmap_rx(priv, mapping); |
272 | ipoib_ud_skb_put_frags(priv, skb, wc->byte_len); | |
2439a6e6 | 273 | |
2439a6e6 RD |
274 | skb_pull(skb, IB_GRH_BYTES); |
275 | ||
1b844afe RD |
276 | skb->protocol = ((struct ipoib_header *) skb->data)->proto; |
277 | skb_reset_mac_header(skb); | |
278 | skb_pull(skb, IPOIB_ENCAP_LEN); | |
279 | ||
de903512 RD |
280 | ++dev->stats.rx_packets; |
281 | dev->stats.rx_bytes += skb->len; | |
1b844afe RD |
282 | |
283 | skb->dev = dev; | |
284 | /* XXX get correct PACKET_ type here */ | |
285 | skb->pkt_type = PACKET_HOST; | |
6046136c EC |
286 | |
287 | if (test_bit(IPOIB_FLAG_CSUM, &priv->flags) && likely(wc->csum_ok)) | |
288 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
289 | ||
af40da89 VS |
290 | if (dev->features & NETIF_F_LRO) |
291 | lro_receive_skb(&priv->lro.lro_mgr, skb, NULL); | |
292 | else | |
293 | netif_receive_skb(skb); | |
1da177e4 | 294 | |
2439a6e6 RD |
295 | repost: |
296 | if (unlikely(ipoib_ib_post_receive(dev, wr_id))) | |
297 | ipoib_warn(priv, "ipoib_ib_post_receive failed " | |
298 | "for buf %d\n", wr_id); | |
299 | } | |
1da177e4 | 300 | |
7143740d EC |
301 | static int ipoib_dma_map_tx(struct ib_device *ca, |
302 | struct ipoib_tx_buf *tx_req) | |
303 | { | |
304 | struct sk_buff *skb = tx_req->skb; | |
305 | u64 *mapping = tx_req->mapping; | |
306 | int i; | |
40ca1988 | 307 | int off; |
7143740d | 308 | |
40ca1988 EC |
309 | if (skb_headlen(skb)) { |
310 | mapping[0] = ib_dma_map_single(ca, skb->data, skb_headlen(skb), | |
311 | DMA_TO_DEVICE); | |
312 | if (unlikely(ib_dma_mapping_error(ca, mapping[0]))) | |
313 | return -EIO; | |
314 | ||
315 | off = 1; | |
316 | } else | |
317 | off = 0; | |
7143740d EC |
318 | |
319 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
320 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
40ca1988 | 321 | mapping[i + off] = ib_dma_map_page(ca, frag->page, |
7143740d EC |
322 | frag->page_offset, frag->size, |
323 | DMA_TO_DEVICE); | |
40ca1988 | 324 | if (unlikely(ib_dma_mapping_error(ca, mapping[i + off]))) |
7143740d EC |
325 | goto partial_error; |
326 | } | |
327 | return 0; | |
328 | ||
329 | partial_error: | |
7143740d EC |
330 | for (; i > 0; --i) { |
331 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; | |
40ca1988 | 332 | ib_dma_unmap_page(ca, mapping[i - !off], frag->size, DMA_TO_DEVICE); |
7143740d | 333 | } |
40ca1988 EC |
334 | |
335 | if (off) | |
336 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
337 | ||
7143740d EC |
338 | return -EIO; |
339 | } | |
340 | ||
341 | static void ipoib_dma_unmap_tx(struct ib_device *ca, | |
342 | struct ipoib_tx_buf *tx_req) | |
343 | { | |
344 | struct sk_buff *skb = tx_req->skb; | |
345 | u64 *mapping = tx_req->mapping; | |
346 | int i; | |
40ca1988 | 347 | int off; |
7143740d | 348 | |
40ca1988 EC |
349 | if (skb_headlen(skb)) { |
350 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
351 | off = 1; | |
352 | } else | |
353 | off = 0; | |
7143740d EC |
354 | |
355 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
356 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
40ca1988 | 357 | ib_dma_unmap_page(ca, mapping[i + off], frag->size, |
7143740d EC |
358 | DMA_TO_DEVICE); |
359 | } | |
360 | } | |
361 | ||
2439a6e6 RD |
362 | static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) |
363 | { | |
364 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
365 | unsigned int wr_id = wc->wr_id; | |
366 | struct ipoib_tx_buf *tx_req; | |
1da177e4 | 367 | |
a89875fc RD |
368 | ipoib_dbg_data(priv, "send completion: id %d, status: %d\n", |
369 | wr_id, wc->status); | |
1da177e4 | 370 | |
2439a6e6 RD |
371 | if (unlikely(wr_id >= ipoib_sendq_size)) { |
372 | ipoib_warn(priv, "send completion event with wrid %d (> %d)\n", | |
373 | wr_id, ipoib_sendq_size); | |
374 | return; | |
1da177e4 | 375 | } |
2439a6e6 RD |
376 | |
377 | tx_req = &priv->tx_ring[wr_id]; | |
378 | ||
7143740d | 379 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
2439a6e6 | 380 | |
de903512 RD |
381 | ++dev->stats.tx_packets; |
382 | dev->stats.tx_bytes += tx_req->skb->len; | |
2439a6e6 RD |
383 | |
384 | dev_kfree_skb_any(tx_req->skb); | |
385 | ||
2439a6e6 | 386 | ++priv->tx_tail; |
1b524963 MT |
387 | if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) && |
388 | netif_queue_stopped(dev) && | |
389 | test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) | |
2439a6e6 | 390 | netif_wake_queue(dev); |
2439a6e6 RD |
391 | |
392 | if (wc->status != IB_WC_SUCCESS && | |
393 | wc->status != IB_WC_WR_FLUSH_ERR) | |
394 | ipoib_warn(priv, "failed send event " | |
395 | "(status=%d, wrid=%d vend_err %x)\n", | |
396 | wc->status, wr_id, wc->vendor_err); | |
397 | } | |
398 | ||
f56bcd80 EC |
399 | static int poll_tx(struct ipoib_dev_priv *priv) |
400 | { | |
401 | int n, i; | |
402 | ||
403 | n = ib_poll_cq(priv->send_cq, MAX_SEND_CQE, priv->send_wc); | |
404 | for (i = 0; i < n; ++i) | |
405 | ipoib_ib_handle_tx_wc(priv->dev, priv->send_wc + i); | |
406 | ||
407 | return n == MAX_SEND_CQE; | |
408 | } | |
409 | ||
bea3348e | 410 | int ipoib_poll(struct napi_struct *napi, int budget) |
2439a6e6 | 411 | { |
bea3348e SH |
412 | struct ipoib_dev_priv *priv = container_of(napi, struct ipoib_dev_priv, napi); |
413 | struct net_device *dev = priv->dev; | |
8d1cc86a RD |
414 | int done; |
415 | int t; | |
8d1cc86a RD |
416 | int n, i; |
417 | ||
418 | done = 0; | |
8d1cc86a | 419 | |
bea3348e SH |
420 | poll_more: |
421 | while (done < budget) { | |
422 | int max = (budget - done); | |
423 | ||
8d1cc86a | 424 | t = min(IPOIB_NUM_WC, max); |
f56bcd80 | 425 | n = ib_poll_cq(priv->recv_cq, t, priv->ibwc); |
8d1cc86a | 426 | |
bea3348e | 427 | for (i = 0; i < n; i++) { |
8d1cc86a RD |
428 | struct ib_wc *wc = priv->ibwc + i; |
429 | ||
1b524963 | 430 | if (wc->wr_id & IPOIB_OP_RECV) { |
8d1cc86a | 431 | ++done; |
1b524963 MT |
432 | if (wc->wr_id & IPOIB_OP_CM) |
433 | ipoib_cm_handle_rx_wc(dev, wc); | |
434 | else | |
435 | ipoib_ib_handle_rx_wc(dev, wc); | |
f56bcd80 EC |
436 | } else |
437 | ipoib_cm_handle_tx_wc(priv->dev, wc); | |
8d1cc86a RD |
438 | } |
439 | ||
bea3348e | 440 | if (n != t) |
8d1cc86a | 441 | break; |
8d1cc86a RD |
442 | } |
443 | ||
bea3348e | 444 | if (done < budget) { |
af40da89 VS |
445 | if (dev->features & NETIF_F_LRO) |
446 | lro_flush_all(&priv->lro.lro_mgr); | |
447 | ||
288379f0 | 448 | napi_complete(napi); |
f56bcd80 | 449 | if (unlikely(ib_req_notify_cq(priv->recv_cq, |
8d1cc86a RD |
450 | IB_CQ_NEXT_COMP | |
451 | IB_CQ_REPORT_MISSED_EVENTS)) && | |
288379f0 | 452 | napi_reschedule(napi)) |
bea3348e | 453 | goto poll_more; |
8d1cc86a RD |
454 | } |
455 | ||
bea3348e | 456 | return done; |
1da177e4 LT |
457 | } |
458 | ||
459 | void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr) | |
460 | { | |
bea3348e SH |
461 | struct net_device *dev = dev_ptr; |
462 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
463 | ||
288379f0 | 464 | napi_schedule(&priv->napi); |
1da177e4 LT |
465 | } |
466 | ||
57ce41d1 EC |
467 | static void drain_tx_cq(struct net_device *dev) |
468 | { | |
469 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
57ce41d1 | 470 | |
943c246e | 471 | netif_tx_lock(dev); |
57ce41d1 EC |
472 | while (poll_tx(priv)) |
473 | ; /* nothing */ | |
474 | ||
475 | if (netif_queue_stopped(dev)) | |
476 | mod_timer(&priv->poll_timer, jiffies + 1); | |
477 | ||
943c246e | 478 | netif_tx_unlock(dev); |
57ce41d1 EC |
479 | } |
480 | ||
481 | void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr) | |
482 | { | |
943c246e RD |
483 | struct ipoib_dev_priv *priv = netdev_priv(dev_ptr); |
484 | ||
485 | mod_timer(&priv->poll_timer, jiffies); | |
57ce41d1 EC |
486 | } |
487 | ||
1da177e4 LT |
488 | static inline int post_send(struct ipoib_dev_priv *priv, |
489 | unsigned int wr_id, | |
490 | struct ib_ah *address, u32 qpn, | |
40ca1988 EC |
491 | struct ipoib_tx_buf *tx_req, |
492 | void *head, int hlen) | |
1da177e4 LT |
493 | { |
494 | struct ib_send_wr *bad_wr; | |
40ca1988 EC |
495 | int i, off; |
496 | struct sk_buff *skb = tx_req->skb; | |
497 | skb_frag_t *frags = skb_shinfo(skb)->frags; | |
498 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
499 | u64 *mapping = tx_req->mapping; | |
500 | ||
501 | if (skb_headlen(skb)) { | |
502 | priv->tx_sge[0].addr = mapping[0]; | |
503 | priv->tx_sge[0].length = skb_headlen(skb); | |
504 | off = 1; | |
505 | } else | |
506 | off = 0; | |
1da177e4 | 507 | |
7143740d | 508 | for (i = 0; i < nr_frags; ++i) { |
40ca1988 EC |
509 | priv->tx_sge[i + off].addr = mapping[i + off]; |
510 | priv->tx_sge[i + off].length = frags[i].size; | |
7143740d | 511 | } |
40ca1988 | 512 | priv->tx_wr.num_sge = nr_frags + off; |
7143740d EC |
513 | priv->tx_wr.wr_id = wr_id; |
514 | priv->tx_wr.wr.ud.remote_qpn = qpn; | |
515 | priv->tx_wr.wr.ud.ah = address; | |
1da177e4 | 516 | |
40ca1988 EC |
517 | if (head) { |
518 | priv->tx_wr.wr.ud.mss = skb_shinfo(skb)->gso_size; | |
519 | priv->tx_wr.wr.ud.header = head; | |
520 | priv->tx_wr.wr.ud.hlen = hlen; | |
521 | priv->tx_wr.opcode = IB_WR_LSO; | |
522 | } else | |
523 | priv->tx_wr.opcode = IB_WR_SEND; | |
524 | ||
1da177e4 LT |
525 | return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr); |
526 | } | |
527 | ||
528 | void ipoib_send(struct net_device *dev, struct sk_buff *skb, | |
529 | struct ipoib_ah *address, u32 qpn) | |
530 | { | |
531 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1993d683 | 532 | struct ipoib_tx_buf *tx_req; |
a48f509b | 533 | int hlen, rc; |
40ca1988 EC |
534 | void *phead; |
535 | ||
536 | if (skb_is_gso(skb)) { | |
537 | hlen = skb_transport_offset(skb) + tcp_hdrlen(skb); | |
538 | phead = skb->data; | |
539 | if (unlikely(!skb_pull(skb, hlen))) { | |
540 | ipoib_warn(priv, "linear data too small\n"); | |
541 | ++dev->stats.tx_dropped; | |
542 | ++dev->stats.tx_errors; | |
543 | dev_kfree_skb_any(skb); | |
544 | return; | |
545 | } | |
546 | } else { | |
547 | if (unlikely(skb->len > priv->mcast_mtu + IPOIB_ENCAP_LEN)) { | |
548 | ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n", | |
549 | skb->len, priv->mcast_mtu + IPOIB_ENCAP_LEN); | |
550 | ++dev->stats.tx_dropped; | |
551 | ++dev->stats.tx_errors; | |
552 | ipoib_cm_skb_too_long(dev, skb, priv->mcast_mtu); | |
553 | return; | |
554 | } | |
555 | phead = NULL; | |
556 | hlen = 0; | |
1da177e4 LT |
557 | } |
558 | ||
559 | ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n", | |
560 | skb->len, address, qpn); | |
561 | ||
562 | /* | |
563 | * We put the skb into the tx_ring _before_ we call post_send() | |
564 | * because it's entirely possible that the completion handler will | |
565 | * run before we execute anything after the post_send(). That | |
566 | * means we have to make sure everything is properly recorded and | |
567 | * our state is consistent before we call post_send(). | |
568 | */ | |
0f485251 | 569 | tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)]; |
1da177e4 | 570 | tx_req->skb = skb; |
7143740d | 571 | if (unlikely(ipoib_dma_map_tx(priv->ca, tx_req))) { |
de903512 | 572 | ++dev->stats.tx_errors; |
73fbe8be RD |
573 | dev_kfree_skb_any(skb); |
574 | return; | |
575 | } | |
1da177e4 | 576 | |
6046136c EC |
577 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
578 | priv->tx_wr.send_flags |= IB_SEND_IP_CSUM; | |
579 | else | |
580 | priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM; | |
581 | ||
57ce41d1 EC |
582 | if (++priv->tx_outstanding == ipoib_sendq_size) { |
583 | ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n"); | |
584 | if (ib_req_notify_cq(priv->send_cq, IB_CQ_NEXT_COMP)) | |
585 | ipoib_warn(priv, "request notify on send CQ failed\n"); | |
586 | netif_stop_queue(dev); | |
587 | } | |
588 | ||
a48f509b OG |
589 | rc = post_send(priv, priv->tx_head & (ipoib_sendq_size - 1), |
590 | address->ah, qpn, tx_req, phead, hlen); | |
591 | if (unlikely(rc)) { | |
592 | ipoib_warn(priv, "post_send failed, error %d\n", rc); | |
de903512 | 593 | ++dev->stats.tx_errors; |
57ce41d1 | 594 | --priv->tx_outstanding; |
7143740d | 595 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
1da177e4 | 596 | dev_kfree_skb_any(skb); |
57ce41d1 EC |
597 | if (netif_queue_stopped(dev)) |
598 | netif_wake_queue(dev); | |
1da177e4 LT |
599 | } else { |
600 | dev->trans_start = jiffies; | |
601 | ||
602 | address->last_send = priv->tx_head; | |
603 | ++priv->tx_head; | |
f56bcd80 | 604 | skb_orphan(skb); |
1da177e4 | 605 | |
1da177e4 | 606 | } |
f56bcd80 EC |
607 | |
608 | if (unlikely(priv->tx_outstanding > MAX_SEND_CQE)) | |
57ce41d1 EC |
609 | while (poll_tx(priv)) |
610 | ; /* nothing */ | |
1da177e4 LT |
611 | } |
612 | ||
613 | static void __ipoib_reap_ah(struct net_device *dev) | |
614 | { | |
615 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
616 | struct ipoib_ah *ah, *tah; | |
617 | LIST_HEAD(remove_list); | |
943c246e RD |
618 | unsigned long flags; |
619 | ||
620 | netif_tx_lock_bh(dev); | |
621 | spin_lock_irqsave(&priv->lock, flags); | |
1da177e4 | 622 | |
1da177e4 | 623 | list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list) |
2181858b | 624 | if ((int) priv->tx_tail - (int) ah->last_send >= 0) { |
1da177e4 | 625 | list_del(&ah->list); |
31c02e21 RD |
626 | ib_destroy_ah(ah->ah); |
627 | kfree(ah); | |
1da177e4 | 628 | } |
943c246e RD |
629 | |
630 | spin_unlock_irqrestore(&priv->lock, flags); | |
631 | netif_tx_unlock_bh(dev); | |
1da177e4 LT |
632 | } |
633 | ||
c4028958 | 634 | void ipoib_reap_ah(struct work_struct *work) |
1da177e4 | 635 | { |
c4028958 DH |
636 | struct ipoib_dev_priv *priv = |
637 | container_of(work, struct ipoib_dev_priv, ah_reap_task.work); | |
638 | struct net_device *dev = priv->dev; | |
1da177e4 LT |
639 | |
640 | __ipoib_reap_ah(dev); | |
641 | ||
642 | if (!test_bit(IPOIB_STOP_REAPER, &priv->flags)) | |
69fc507a AB |
643 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, |
644 | round_jiffies_relative(HZ)); | |
1da177e4 LT |
645 | } |
646 | ||
57ce41d1 EC |
647 | static void ipoib_ib_tx_timer_func(unsigned long ctx) |
648 | { | |
649 | drain_tx_cq((struct net_device *)ctx); | |
650 | } | |
651 | ||
1da177e4 LT |
652 | int ipoib_ib_dev_open(struct net_device *dev) |
653 | { | |
654 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
655 | int ret; | |
656 | ||
26bbf13c YE |
657 | if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &priv->pkey_index)) { |
658 | ipoib_warn(priv, "P_Key 0x%04x not found\n", priv->pkey); | |
659 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
660 | return -1; | |
661 | } | |
662 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
663 | ||
5b6810e0 | 664 | ret = ipoib_init_qp(dev); |
1da177e4 | 665 | if (ret) { |
5b6810e0 | 666 | ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret); |
1da177e4 LT |
667 | return -1; |
668 | } | |
669 | ||
670 | ret = ipoib_ib_post_receives(dev); | |
671 | if (ret) { | |
672 | ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret); | |
26bbf13c | 673 | ipoib_ib_dev_stop(dev, 1); |
1da177e4 LT |
674 | return -1; |
675 | } | |
676 | ||
839fcaba MT |
677 | ret = ipoib_cm_dev_open(dev); |
678 | if (ret) { | |
24bd1e4e | 679 | ipoib_warn(priv, "ipoib_cm_dev_open returned %d\n", ret); |
26bbf13c | 680 | ipoib_ib_dev_stop(dev, 1); |
839fcaba MT |
681 | return -1; |
682 | } | |
683 | ||
1da177e4 | 684 | clear_bit(IPOIB_STOP_REAPER, &priv->flags); |
69fc507a AB |
685 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, |
686 | round_jiffies_relative(HZ)); | |
1da177e4 | 687 | |
e028cc55 YE |
688 | if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
689 | napi_enable(&priv->napi); | |
7a343d4c | 690 | |
1da177e4 LT |
691 | return 0; |
692 | } | |
693 | ||
7a343d4c LA |
694 | static void ipoib_pkey_dev_check_presence(struct net_device *dev) |
695 | { | |
696 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
697 | u16 pkey_index = 0; | |
698 | ||
9fdd5e5b | 699 | if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) |
7a343d4c LA |
700 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); |
701 | else | |
702 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
703 | } | |
704 | ||
1da177e4 LT |
705 | int ipoib_ib_dev_up(struct net_device *dev) |
706 | { | |
707 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
708 | ||
7a343d4c LA |
709 | ipoib_pkey_dev_check_presence(dev); |
710 | ||
711 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
712 | ipoib_dbg(priv, "PKEY is not assigned.\n"); | |
713 | return 0; | |
714 | } | |
715 | ||
1da177e4 LT |
716 | set_bit(IPOIB_FLAG_OPER_UP, &priv->flags); |
717 | ||
718 | return ipoib_mcast_start_thread(dev); | |
719 | } | |
720 | ||
0b3ea082 | 721 | int ipoib_ib_dev_down(struct net_device *dev, int flush) |
1da177e4 LT |
722 | { |
723 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
724 | ||
725 | ipoib_dbg(priv, "downing ib_dev\n"); | |
726 | ||
727 | clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags); | |
728 | netif_carrier_off(dev); | |
729 | ||
730 | /* Shutdown the P_Key thread if still active */ | |
731 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
95ed644f | 732 | mutex_lock(&pkey_mutex); |
1da177e4 | 733 | set_bit(IPOIB_PKEY_STOP, &priv->flags); |
26bbf13c | 734 | cancel_delayed_work(&priv->pkey_poll_task); |
95ed644f | 735 | mutex_unlock(&pkey_mutex); |
0b3ea082 JM |
736 | if (flush) |
737 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
738 | } |
739 | ||
0b3ea082 | 740 | ipoib_mcast_stop_thread(dev, flush); |
1da177e4 LT |
741 | ipoib_mcast_dev_flush(dev); |
742 | ||
1da177e4 LT |
743 | ipoib_flush_paths(dev); |
744 | ||
745 | return 0; | |
746 | } | |
747 | ||
748 | static int recvs_pending(struct net_device *dev) | |
749 | { | |
750 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
751 | int pending = 0; | |
752 | int i; | |
753 | ||
0f485251 | 754 | for (i = 0; i < ipoib_recvq_size; ++i) |
1da177e4 LT |
755 | if (priv->rx_ring[i].skb) |
756 | ++pending; | |
757 | ||
758 | return pending; | |
759 | } | |
760 | ||
2dfbfc37 MT |
761 | void ipoib_drain_cq(struct net_device *dev) |
762 | { | |
763 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
764 | int i, n; | |
943c246e RD |
765 | |
766 | /* | |
767 | * We call completion handling routines that expect to be | |
768 | * called from the BH-disabled NAPI poll context, so disable | |
769 | * BHs here too. | |
770 | */ | |
771 | local_bh_disable(); | |
772 | ||
2dfbfc37 | 773 | do { |
f56bcd80 | 774 | n = ib_poll_cq(priv->recv_cq, IPOIB_NUM_WC, priv->ibwc); |
2dfbfc37 | 775 | for (i = 0; i < n; ++i) { |
ce423ef5 RD |
776 | /* |
777 | * Convert any successful completions to flush | |
778 | * errors to avoid passing packets up the | |
779 | * stack after bringing the device down. | |
780 | */ | |
781 | if (priv->ibwc[i].status == IB_WC_SUCCESS) | |
782 | priv->ibwc[i].status = IB_WC_WR_FLUSH_ERR; | |
783 | ||
1b524963 MT |
784 | if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) { |
785 | if (priv->ibwc[i].wr_id & IPOIB_OP_CM) | |
786 | ipoib_cm_handle_rx_wc(dev, priv->ibwc + i); | |
787 | else | |
788 | ipoib_ib_handle_rx_wc(dev, priv->ibwc + i); | |
f56bcd80 EC |
789 | } else |
790 | ipoib_cm_handle_tx_wc(dev, priv->ibwc + i); | |
2dfbfc37 MT |
791 | } |
792 | } while (n == IPOIB_NUM_WC); | |
f56bcd80 EC |
793 | |
794 | while (poll_tx(priv)) | |
795 | ; /* nothing */ | |
943c246e RD |
796 | |
797 | local_bh_enable(); | |
2dfbfc37 MT |
798 | } |
799 | ||
26bbf13c | 800 | int ipoib_ib_dev_stop(struct net_device *dev, int flush) |
1da177e4 LT |
801 | { |
802 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
803 | struct ib_qp_attr qp_attr; | |
1da177e4 | 804 | unsigned long begin; |
1993d683 | 805 | struct ipoib_tx_buf *tx_req; |
2dfbfc37 | 806 | int i; |
1da177e4 | 807 | |
e028cc55 YE |
808 | if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
809 | napi_disable(&priv->napi); | |
7a343d4c | 810 | |
839fcaba MT |
811 | ipoib_cm_dev_stop(dev); |
812 | ||
3bc12e75 RD |
813 | /* |
814 | * Move our QP to the error state and then reinitialize in | |
815 | * when all work requests have completed or have been flushed. | |
816 | */ | |
1da177e4 | 817 | qp_attr.qp_state = IB_QPS_ERR; |
3bc12e75 | 818 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
819 | ipoib_warn(priv, "Failed to modify QP to ERROR state\n"); |
820 | ||
821 | /* Wait for all sends and receives to complete */ | |
822 | begin = jiffies; | |
823 | ||
824 | while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) { | |
825 | if (time_after(jiffies, begin + 5 * HZ)) { | |
826 | ipoib_warn(priv, "timing out; %d sends %d receives not completed\n", | |
827 | priv->tx_head - priv->tx_tail, recvs_pending(dev)); | |
828 | ||
829 | /* | |
830 | * assume the HW is wedged and just free up | |
831 | * all our pending work requests. | |
832 | */ | |
2181858b | 833 | while ((int) priv->tx_tail - (int) priv->tx_head < 0) { |
1da177e4 | 834 | tx_req = &priv->tx_ring[priv->tx_tail & |
0f485251 | 835 | (ipoib_sendq_size - 1)]; |
7143740d | 836 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
1da177e4 LT |
837 | dev_kfree_skb_any(tx_req->skb); |
838 | ++priv->tx_tail; | |
1b524963 | 839 | --priv->tx_outstanding; |
1da177e4 LT |
840 | } |
841 | ||
37ccf9df RC |
842 | for (i = 0; i < ipoib_recvq_size; ++i) { |
843 | struct ipoib_rx_buf *rx_req; | |
844 | ||
845 | rx_req = &priv->rx_ring[i]; | |
846 | if (!rx_req->skb) | |
847 | continue; | |
bc7b3a36 SM |
848 | ipoib_ud_dma_unmap_rx(priv, |
849 | priv->rx_ring[i].mapping); | |
37ccf9df RC |
850 | dev_kfree_skb_any(rx_req->skb); |
851 | rx_req->skb = NULL; | |
852 | } | |
1da177e4 LT |
853 | |
854 | goto timeout; | |
855 | } | |
856 | ||
2dfbfc37 | 857 | ipoib_drain_cq(dev); |
8d1cc86a | 858 | |
1da177e4 LT |
859 | msleep(1); |
860 | } | |
861 | ||
862 | ipoib_dbg(priv, "All sends and receives done.\n"); | |
863 | ||
864 | timeout: | |
57ce41d1 | 865 | del_timer_sync(&priv->poll_timer); |
1da177e4 | 866 | qp_attr.qp_state = IB_QPS_RESET; |
3bc12e75 | 867 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
868 | ipoib_warn(priv, "Failed to modify QP to RESET state\n"); |
869 | ||
870 | /* Wait for all AHs to be reaped */ | |
871 | set_bit(IPOIB_STOP_REAPER, &priv->flags); | |
872 | cancel_delayed_work(&priv->ah_reap_task); | |
26bbf13c YE |
873 | if (flush) |
874 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
875 | |
876 | begin = jiffies; | |
877 | ||
878 | while (!list_empty(&priv->dead_ahs)) { | |
879 | __ipoib_reap_ah(dev); | |
880 | ||
881 | if (time_after(jiffies, begin + HZ)) { | |
882 | ipoib_warn(priv, "timing out; will leak address handles\n"); | |
883 | break; | |
884 | } | |
885 | ||
886 | msleep(1); | |
887 | } | |
888 | ||
f56bcd80 | 889 | ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP); |
8d1cc86a | 890 | |
1da177e4 LT |
891 | return 0; |
892 | } | |
893 | ||
894 | int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port) | |
895 | { | |
896 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
897 | ||
898 | priv->ca = ca; | |
899 | priv->port = port; | |
900 | priv->qp = NULL; | |
901 | ||
902 | if (ipoib_transport_dev_init(dev, ca)) { | |
903 | printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name); | |
904 | return -ENODEV; | |
905 | } | |
906 | ||
2767840a RD |
907 | setup_timer(&priv->poll_timer, ipoib_ib_tx_timer_func, |
908 | (unsigned long) dev); | |
909 | ||
1da177e4 LT |
910 | if (dev->flags & IFF_UP) { |
911 | if (ipoib_ib_dev_open(dev)) { | |
912 | ipoib_transport_dev_cleanup(dev); | |
913 | return -ENODEV; | |
914 | } | |
915 | } | |
916 | ||
917 | return 0; | |
918 | } | |
919 | ||
ee1e2c82 MS |
920 | static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, |
921 | enum ipoib_flush_level level) | |
1da177e4 | 922 | { |
26bbf13c | 923 | struct ipoib_dev_priv *cpriv; |
c4028958 | 924 | struct net_device *dev = priv->dev; |
26bbf13c YE |
925 | u16 new_index; |
926 | ||
927 | mutex_lock(&priv->vlan_mutex); | |
1da177e4 | 928 | |
26bbf13c YE |
929 | /* |
930 | * Flush any child interfaces too -- they might be up even if | |
931 | * the parent is down. | |
932 | */ | |
933 | list_for_each_entry(cpriv, &priv->child_intfs, list) | |
ee1e2c82 | 934 | __ipoib_ib_dev_flush(cpriv, level); |
26bbf13c YE |
935 | |
936 | mutex_unlock(&priv->vlan_mutex); | |
937 | ||
938 | if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) { | |
7a343d4c LA |
939 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n"); |
940 | return; | |
941 | } | |
942 | ||
943 | if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { | |
944 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n"); | |
1da177e4 | 945 | return; |
7a343d4c | 946 | } |
1da177e4 | 947 | |
ee1e2c82 | 948 | if (level == IPOIB_FLUSH_HEAVY) { |
26bbf13c YE |
949 | if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &new_index)) { |
950 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
951 | ipoib_ib_dev_down(dev, 0); | |
167c4265 | 952 | ipoib_ib_dev_stop(dev, 0); |
9fdd5e5b RD |
953 | if (ipoib_pkey_dev_delay_open(dev)) |
954 | return; | |
26bbf13c | 955 | } |
26bbf13c YE |
956 | |
957 | /* restart QP only if P_Key index is changed */ | |
9fdd5e5b RD |
958 | if (test_and_set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags) && |
959 | new_index == priv->pkey_index) { | |
26bbf13c YE |
960 | ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n"); |
961 | return; | |
962 | } | |
963 | priv->pkey_index = new_index; | |
964 | } | |
965 | ||
ee1e2c82 MS |
966 | if (level == IPOIB_FLUSH_LIGHT) { |
967 | ipoib_mark_paths_invalid(dev); | |
968 | ipoib_mcast_dev_flush(dev); | |
969 | } | |
1da177e4 | 970 | |
ee1e2c82 MS |
971 | if (level >= IPOIB_FLUSH_NORMAL) |
972 | ipoib_ib_dev_down(dev, 0); | |
1da177e4 | 973 | |
ee1e2c82 | 974 | if (level == IPOIB_FLUSH_HEAVY) { |
26bbf13c YE |
975 | ipoib_ib_dev_stop(dev, 0); |
976 | ipoib_ib_dev_open(dev); | |
977 | } | |
978 | ||
1da177e4 LT |
979 | /* |
980 | * The device could have been brought down between the start and when | |
981 | * we get here, don't bring it back up if it's not configured up | |
982 | */ | |
5ccd0255 | 983 | if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { |
ee1e2c82 MS |
984 | if (level >= IPOIB_FLUSH_NORMAL) |
985 | ipoib_ib_dev_up(dev); | |
c4028958 | 986 | ipoib_mcast_restart_task(&priv->restart_task); |
5ccd0255 | 987 | } |
26bbf13c | 988 | } |
1da177e4 | 989 | |
ee1e2c82 MS |
990 | void ipoib_ib_dev_flush_light(struct work_struct *work) |
991 | { | |
992 | struct ipoib_dev_priv *priv = | |
993 | container_of(work, struct ipoib_dev_priv, flush_light); | |
994 | ||
995 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_LIGHT); | |
996 | } | |
997 | ||
998 | void ipoib_ib_dev_flush_normal(struct work_struct *work) | |
26bbf13c YE |
999 | { |
1000 | struct ipoib_dev_priv *priv = | |
ee1e2c82 | 1001 | container_of(work, struct ipoib_dev_priv, flush_normal); |
4f71055a | 1002 | |
ee1e2c82 | 1003 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_NORMAL); |
26bbf13c | 1004 | } |
4f71055a | 1005 | |
ee1e2c82 | 1006 | void ipoib_ib_dev_flush_heavy(struct work_struct *work) |
26bbf13c YE |
1007 | { |
1008 | struct ipoib_dev_priv *priv = | |
ee1e2c82 | 1009 | container_of(work, struct ipoib_dev_priv, flush_heavy); |
26bbf13c | 1010 | |
ee1e2c82 | 1011 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_HEAVY); |
1da177e4 LT |
1012 | } |
1013 | ||
1014 | void ipoib_ib_dev_cleanup(struct net_device *dev) | |
1015 | { | |
1016 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1017 | ||
1018 | ipoib_dbg(priv, "cleaning up ib_dev\n"); | |
1019 | ||
8d2cae06 | 1020 | ipoib_mcast_stop_thread(dev, 1); |
988bd503 | 1021 | ipoib_mcast_dev_flush(dev); |
1da177e4 LT |
1022 | |
1023 | ipoib_transport_dev_cleanup(dev); | |
1024 | } | |
1025 | ||
1026 | /* | |
1027 | * Delayed P_Key Assigment Interim Support | |
1028 | * | |
1029 | * The following is initial implementation of delayed P_Key assigment | |
1030 | * mechanism. It is using the same approach implemented for the multicast | |
1031 | * group join. The single goal of this implementation is to quickly address | |
1032 | * Bug #2507. This implementation will probably be removed when the P_Key | |
1033 | * change async notification is available. | |
1034 | */ | |
1da177e4 | 1035 | |
c4028958 | 1036 | void ipoib_pkey_poll(struct work_struct *work) |
1da177e4 | 1037 | { |
c4028958 | 1038 | struct ipoib_dev_priv *priv = |
26bbf13c | 1039 | container_of(work, struct ipoib_dev_priv, pkey_poll_task.work); |
c4028958 | 1040 | struct net_device *dev = priv->dev; |
1da177e4 LT |
1041 | |
1042 | ipoib_pkey_dev_check_presence(dev); | |
1043 | ||
1044 | if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) | |
1045 | ipoib_open(dev); | |
1046 | else { | |
95ed644f | 1047 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
1048 | if (!test_bit(IPOIB_PKEY_STOP, &priv->flags)) |
1049 | queue_delayed_work(ipoib_workqueue, | |
26bbf13c | 1050 | &priv->pkey_poll_task, |
1da177e4 | 1051 | HZ); |
95ed644f | 1052 | mutex_unlock(&pkey_mutex); |
1da177e4 LT |
1053 | } |
1054 | } | |
1055 | ||
1056 | int ipoib_pkey_dev_delay_open(struct net_device *dev) | |
1057 | { | |
1058 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1059 | ||
1060 | /* Look for the interface pkey value in the IB Port P_Key table and */ | |
1061 | /* set the interface pkey assigment flag */ | |
1062 | ipoib_pkey_dev_check_presence(dev); | |
1063 | ||
1064 | /* P_Key value not assigned yet - start polling */ | |
1065 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
95ed644f | 1066 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
1067 | clear_bit(IPOIB_PKEY_STOP, &priv->flags); |
1068 | queue_delayed_work(ipoib_workqueue, | |
26bbf13c | 1069 | &priv->pkey_poll_task, |
1da177e4 | 1070 | HZ); |
95ed644f | 1071 | mutex_unlock(&pkey_mutex); |
1da177e4 LT |
1072 | return 1; |
1073 | } | |
1074 | ||
1075 | return 0; | |
1076 | } |