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> | |
fec14d2f | 37 | #include <linux/moduleparam.h> |
1da177e4 | 38 | #include <linux/dma-mapping.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
1da177e4 | 40 | |
40ca1988 EC |
41 | #include <linux/ip.h> |
42 | #include <linux/tcp.h> | |
1da177e4 LT |
43 | |
44 | #include "ipoib.h" | |
45 | ||
46 | #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA | |
47 | static int data_debug_level; | |
48 | ||
49 | module_param(data_debug_level, int, 0644); | |
50 | MODULE_PARM_DESC(data_debug_level, | |
51 | "Enable data path debug tracing if > 0"); | |
52 | #endif | |
53 | ||
95ed644f | 54 | static DEFINE_MUTEX(pkey_mutex); |
1da177e4 LT |
55 | |
56 | struct ipoib_ah *ipoib_create_ah(struct net_device *dev, | |
57 | struct ib_pd *pd, struct ib_ah_attr *attr) | |
58 | { | |
59 | struct ipoib_ah *ah; | |
3874397c | 60 | struct ib_ah *vah; |
1da177e4 LT |
61 | |
62 | ah = kmalloc(sizeof *ah, GFP_KERNEL); | |
63 | if (!ah) | |
3874397c | 64 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
65 | |
66 | ah->dev = dev; | |
67 | ah->last_send = 0; | |
68 | kref_init(&ah->ref); | |
69 | ||
3874397c MM |
70 | vah = ib_create_ah(pd, attr); |
71 | if (IS_ERR(vah)) { | |
1da177e4 | 72 | kfree(ah); |
3874397c MM |
73 | ah = (struct ipoib_ah *)vah; |
74 | } else { | |
75 | ah->ah = vah; | |
1da177e4 | 76 | ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah); |
3874397c | 77 | } |
1da177e4 LT |
78 | |
79 | return ah; | |
80 | } | |
81 | ||
82 | void ipoib_free_ah(struct kref *kref) | |
83 | { | |
84 | struct ipoib_ah *ah = container_of(kref, struct ipoib_ah, ref); | |
85 | struct ipoib_dev_priv *priv = netdev_priv(ah->dev); | |
86 | ||
87 | unsigned long flags; | |
88 | ||
31c02e21 RD |
89 | spin_lock_irqsave(&priv->lock, flags); |
90 | list_add_tail(&ah->list, &priv->dead_ahs); | |
91 | spin_unlock_irqrestore(&priv->lock, flags); | |
1da177e4 LT |
92 | } |
93 | ||
bc7b3a36 SM |
94 | static void ipoib_ud_dma_unmap_rx(struct ipoib_dev_priv *priv, |
95 | u64 mapping[IPOIB_UD_RX_SG]) | |
96 | { | |
97 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) { | |
98 | ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_UD_HEAD_SIZE, | |
99 | DMA_FROM_DEVICE); | |
100 | ib_dma_unmap_page(priv->ca, mapping[1], PAGE_SIZE, | |
101 | DMA_FROM_DEVICE); | |
102 | } else | |
103 | ib_dma_unmap_single(priv->ca, mapping[0], | |
104 | IPOIB_UD_BUF_SIZE(priv->max_ib_mtu), | |
105 | DMA_FROM_DEVICE); | |
106 | } | |
107 | ||
108 | static void ipoib_ud_skb_put_frags(struct ipoib_dev_priv *priv, | |
109 | struct sk_buff *skb, | |
110 | unsigned int length) | |
111 | { | |
112 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) { | |
113 | skb_frag_t *frag = &skb_shinfo(skb)->frags[0]; | |
114 | unsigned int size; | |
115 | /* | |
116 | * There is only two buffers needed for max_payload = 4K, | |
117 | * first buf size is IPOIB_UD_HEAD_SIZE | |
118 | */ | |
119 | skb->tail += IPOIB_UD_HEAD_SIZE; | |
120 | skb->len += length; | |
121 | ||
122 | size = length - IPOIB_UD_HEAD_SIZE; | |
123 | ||
9e903e08 | 124 | skb_frag_size_set(frag, size); |
bc7b3a36 SM |
125 | skb->data_len += size; |
126 | skb->truesize += size; | |
127 | } else | |
128 | skb_put(skb, length); | |
129 | ||
130 | } | |
131 | ||
1993d683 | 132 | static int ipoib_ib_post_receive(struct net_device *dev, int id) |
1da177e4 | 133 | { |
1993d683 | 134 | struct ipoib_dev_priv *priv = netdev_priv(dev); |
1da177e4 | 135 | struct ib_recv_wr *bad_wr; |
1993d683 RD |
136 | int ret; |
137 | ||
bc7b3a36 SM |
138 | priv->rx_wr.wr_id = id | IPOIB_OP_RECV; |
139 | priv->rx_sge[0].addr = priv->rx_ring[id].mapping[0]; | |
140 | priv->rx_sge[1].addr = priv->rx_ring[id].mapping[1]; | |
1993d683 | 141 | |
1993d683 | 142 | |
bc7b3a36 | 143 | ret = ib_post_recv(priv->qp, &priv->rx_wr, &bad_wr); |
1993d683 RD |
144 | if (unlikely(ret)) { |
145 | ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret); | |
bc7b3a36 | 146 | ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[id].mapping); |
1993d683 RD |
147 | dev_kfree_skb_any(priv->rx_ring[id].skb); |
148 | priv->rx_ring[id].skb = NULL; | |
149 | } | |
1da177e4 | 150 | |
1993d683 | 151 | return ret; |
1da177e4 LT |
152 | } |
153 | ||
bc7b3a36 | 154 | static struct sk_buff *ipoib_alloc_rx_skb(struct net_device *dev, int id) |
1da177e4 LT |
155 | { |
156 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
157 | struct sk_buff *skb; | |
bc7b3a36 SM |
158 | int buf_size; |
159 | u64 *mapping; | |
1da177e4 | 160 | |
bc7b3a36 SM |
161 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) |
162 | buf_size = IPOIB_UD_HEAD_SIZE; | |
163 | else | |
164 | buf_size = IPOIB_UD_BUF_SIZE(priv->max_ib_mtu); | |
165 | ||
166 | skb = dev_alloc_skb(buf_size + 4); | |
167 | if (unlikely(!skb)) | |
168 | return NULL; | |
1993d683 RD |
169 | |
170 | /* | |
171 | * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte | |
172 | * header. So we need 4 more bytes to get to 48 and align the | |
173 | * IP header to a multiple of 16. | |
174 | */ | |
175 | skb_reserve(skb, 4); | |
176 | ||
bc7b3a36 SM |
177 | mapping = priv->rx_ring[id].mapping; |
178 | mapping[0] = ib_dma_map_single(priv->ca, skb->data, buf_size, | |
179 | DMA_FROM_DEVICE); | |
180 | if (unlikely(ib_dma_mapping_error(priv->ca, mapping[0]))) | |
181 | goto error; | |
182 | ||
183 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) { | |
184 | struct page *page = alloc_page(GFP_ATOMIC); | |
185 | if (!page) | |
186 | goto partial_error; | |
187 | skb_fill_page_desc(skb, 0, page, 0, PAGE_SIZE); | |
188 | mapping[1] = | |
5581be3b | 189 | ib_dma_map_page(priv->ca, page, |
bc7b3a36 SM |
190 | 0, PAGE_SIZE, DMA_FROM_DEVICE); |
191 | if (unlikely(ib_dma_mapping_error(priv->ca, mapping[1]))) | |
192 | goto partial_error; | |
1da177e4 LT |
193 | } |
194 | ||
bc7b3a36 SM |
195 | priv->rx_ring[id].skb = skb; |
196 | return skb; | |
1993d683 | 197 | |
bc7b3a36 SM |
198 | partial_error: |
199 | ib_dma_unmap_single(priv->ca, mapping[0], buf_size, DMA_FROM_DEVICE); | |
200 | error: | |
201 | dev_kfree_skb_any(skb); | |
202 | return NULL; | |
1da177e4 LT |
203 | } |
204 | ||
205 | static int ipoib_ib_post_receives(struct net_device *dev) | |
206 | { | |
207 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
208 | int i; | |
209 | ||
0f485251 | 210 | for (i = 0; i < ipoib_recvq_size; ++i) { |
bc7b3a36 | 211 | if (!ipoib_alloc_rx_skb(dev, i)) { |
1993d683 RD |
212 | ipoib_warn(priv, "failed to allocate receive buffer %d\n", i); |
213 | return -ENOMEM; | |
214 | } | |
1da177e4 LT |
215 | if (ipoib_ib_post_receive(dev, i)) { |
216 | ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i); | |
217 | return -EIO; | |
218 | } | |
219 | } | |
220 | ||
221 | return 0; | |
222 | } | |
223 | ||
2439a6e6 | 224 | static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) |
1da177e4 LT |
225 | { |
226 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
2439a6e6 RD |
227 | unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV; |
228 | struct sk_buff *skb; | |
bc7b3a36 | 229 | u64 mapping[IPOIB_UD_RX_SG]; |
fed1db33 | 230 | union ib_gid *dgid; |
1da177e4 | 231 | |
a89875fc RD |
232 | ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n", |
233 | wr_id, wc->status); | |
1da177e4 | 234 | |
2439a6e6 RD |
235 | if (unlikely(wr_id >= ipoib_recvq_size)) { |
236 | ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n", | |
237 | wr_id, ipoib_recvq_size); | |
238 | return; | |
239 | } | |
240 | ||
241 | skb = priv->rx_ring[wr_id].skb; | |
2439a6e6 RD |
242 | |
243 | if (unlikely(wc->status != IB_WC_SUCCESS)) { | |
244 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
245 | ipoib_warn(priv, "failed recv event " | |
246 | "(status=%d, wrid=%d vend_err %x)\n", | |
247 | wc->status, wr_id, wc->vendor_err); | |
bc7b3a36 | 248 | ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[wr_id].mapping); |
2439a6e6 RD |
249 | dev_kfree_skb_any(skb); |
250 | priv->rx_ring[wr_id].skb = NULL; | |
251 | return; | |
252 | } | |
1da177e4 | 253 | |
1b844afe RD |
254 | /* |
255 | * Drop packets that this interface sent, ie multicast packets | |
256 | * that the HCA has replicated. | |
257 | */ | |
258 | if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num) | |
259 | goto repost; | |
260 | ||
bc7b3a36 SM |
261 | memcpy(mapping, priv->rx_ring[wr_id].mapping, |
262 | IPOIB_UD_RX_SG * sizeof *mapping); | |
263 | ||
2439a6e6 RD |
264 | /* |
265 | * If we can't allocate a new RX buffer, dump | |
266 | * this packet and reuse the old buffer. | |
267 | */ | |
bc7b3a36 | 268 | if (unlikely(!ipoib_alloc_rx_skb(dev, wr_id))) { |
de903512 | 269 | ++dev->stats.rx_dropped; |
2439a6e6 RD |
270 | goto repost; |
271 | } | |
272 | ||
273 | ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n", | |
274 | wc->byte_len, wc->slid); | |
275 | ||
bc7b3a36 SM |
276 | ipoib_ud_dma_unmap_rx(priv, mapping); |
277 | ipoib_ud_skb_put_frags(priv, skb, wc->byte_len); | |
2439a6e6 | 278 | |
fed1db33 CL |
279 | /* First byte of dgid signals multicast when 0xff */ |
280 | dgid = &((struct ib_grh *)skb->data)->dgid; | |
281 | ||
282 | if (!(wc->wc_flags & IB_WC_GRH) || dgid->raw[0] != 0xff) | |
283 | skb->pkt_type = PACKET_HOST; | |
284 | else if (memcmp(dgid, dev->broadcast + 4, sizeof(union ib_gid)) == 0) | |
285 | skb->pkt_type = PACKET_BROADCAST; | |
286 | else | |
287 | skb->pkt_type = PACKET_MULTICAST; | |
288 | ||
2439a6e6 RD |
289 | skb_pull(skb, IB_GRH_BYTES); |
290 | ||
1b844afe RD |
291 | skb->protocol = ((struct ipoib_header *) skb->data)->proto; |
292 | skb_reset_mac_header(skb); | |
293 | skb_pull(skb, IPOIB_ENCAP_LEN); | |
294 | ||
de903512 RD |
295 | ++dev->stats.rx_packets; |
296 | dev->stats.rx_bytes += skb->len; | |
1b844afe RD |
297 | |
298 | skb->dev = dev; | |
3d96c74d | 299 | if ((dev->features & NETIF_F_RXCSUM) && likely(wc->csum_ok)) |
6046136c EC |
300 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
301 | ||
8ae31e5b | 302 | napi_gro_receive(&priv->napi, skb); |
1da177e4 | 303 | |
2439a6e6 RD |
304 | repost: |
305 | if (unlikely(ipoib_ib_post_receive(dev, wr_id))) | |
306 | ipoib_warn(priv, "ipoib_ib_post_receive failed " | |
307 | "for buf %d\n", wr_id); | |
308 | } | |
1da177e4 | 309 | |
7143740d EC |
310 | static int ipoib_dma_map_tx(struct ib_device *ca, |
311 | struct ipoib_tx_buf *tx_req) | |
312 | { | |
313 | struct sk_buff *skb = tx_req->skb; | |
314 | u64 *mapping = tx_req->mapping; | |
315 | int i; | |
40ca1988 | 316 | int off; |
7143740d | 317 | |
40ca1988 EC |
318 | if (skb_headlen(skb)) { |
319 | mapping[0] = ib_dma_map_single(ca, skb->data, skb_headlen(skb), | |
320 | DMA_TO_DEVICE); | |
321 | if (unlikely(ib_dma_mapping_error(ca, mapping[0]))) | |
322 | return -EIO; | |
323 | ||
324 | off = 1; | |
325 | } else | |
326 | off = 0; | |
7143740d EC |
327 | |
328 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
9e903e08 | 329 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
5581be3b IC |
330 | mapping[i + off] = ib_dma_map_page(ca, |
331 | skb_frag_page(frag), | |
9e903e08 | 332 | frag->page_offset, skb_frag_size(frag), |
7143740d | 333 | DMA_TO_DEVICE); |
40ca1988 | 334 | if (unlikely(ib_dma_mapping_error(ca, mapping[i + off]))) |
7143740d EC |
335 | goto partial_error; |
336 | } | |
337 | return 0; | |
338 | ||
339 | partial_error: | |
7143740d | 340 | for (; i > 0; --i) { |
9e903e08 ED |
341 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; |
342 | ||
343 | ib_dma_unmap_page(ca, mapping[i - !off], skb_frag_size(frag), DMA_TO_DEVICE); | |
7143740d | 344 | } |
40ca1988 EC |
345 | |
346 | if (off) | |
347 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
348 | ||
7143740d EC |
349 | return -EIO; |
350 | } | |
351 | ||
352 | static void ipoib_dma_unmap_tx(struct ib_device *ca, | |
353 | struct ipoib_tx_buf *tx_req) | |
354 | { | |
355 | struct sk_buff *skb = tx_req->skb; | |
356 | u64 *mapping = tx_req->mapping; | |
357 | int i; | |
40ca1988 | 358 | int off; |
7143740d | 359 | |
40ca1988 EC |
360 | if (skb_headlen(skb)) { |
361 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
362 | off = 1; | |
363 | } else | |
364 | off = 0; | |
7143740d EC |
365 | |
366 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
9e903e08 ED |
367 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
368 | ||
369 | ib_dma_unmap_page(ca, mapping[i + off], skb_frag_size(frag), | |
7143740d EC |
370 | DMA_TO_DEVICE); |
371 | } | |
372 | } | |
373 | ||
2439a6e6 RD |
374 | static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) |
375 | { | |
376 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
377 | unsigned int wr_id = wc->wr_id; | |
378 | struct ipoib_tx_buf *tx_req; | |
1da177e4 | 379 | |
a89875fc RD |
380 | ipoib_dbg_data(priv, "send completion: id %d, status: %d\n", |
381 | wr_id, wc->status); | |
1da177e4 | 382 | |
2439a6e6 RD |
383 | if (unlikely(wr_id >= ipoib_sendq_size)) { |
384 | ipoib_warn(priv, "send completion event with wrid %d (> %d)\n", | |
385 | wr_id, ipoib_sendq_size); | |
386 | return; | |
1da177e4 | 387 | } |
2439a6e6 RD |
388 | |
389 | tx_req = &priv->tx_ring[wr_id]; | |
390 | ||
7143740d | 391 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
2439a6e6 | 392 | |
de903512 RD |
393 | ++dev->stats.tx_packets; |
394 | dev->stats.tx_bytes += tx_req->skb->len; | |
2439a6e6 RD |
395 | |
396 | dev_kfree_skb_any(tx_req->skb); | |
397 | ||
2439a6e6 | 398 | ++priv->tx_tail; |
1b524963 MT |
399 | if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) && |
400 | netif_queue_stopped(dev) && | |
401 | test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) | |
2439a6e6 | 402 | netif_wake_queue(dev); |
2439a6e6 RD |
403 | |
404 | if (wc->status != IB_WC_SUCCESS && | |
405 | wc->status != IB_WC_WR_FLUSH_ERR) | |
406 | ipoib_warn(priv, "failed send event " | |
407 | "(status=%d, wrid=%d vend_err %x)\n", | |
408 | wc->status, wr_id, wc->vendor_err); | |
409 | } | |
410 | ||
f56bcd80 EC |
411 | static int poll_tx(struct ipoib_dev_priv *priv) |
412 | { | |
413 | int n, i; | |
414 | ||
415 | n = ib_poll_cq(priv->send_cq, MAX_SEND_CQE, priv->send_wc); | |
416 | for (i = 0; i < n; ++i) | |
417 | ipoib_ib_handle_tx_wc(priv->dev, priv->send_wc + i); | |
418 | ||
419 | return n == MAX_SEND_CQE; | |
420 | } | |
421 | ||
bea3348e | 422 | int ipoib_poll(struct napi_struct *napi, int budget) |
2439a6e6 | 423 | { |
bea3348e SH |
424 | struct ipoib_dev_priv *priv = container_of(napi, struct ipoib_dev_priv, napi); |
425 | struct net_device *dev = priv->dev; | |
8d1cc86a RD |
426 | int done; |
427 | int t; | |
8d1cc86a RD |
428 | int n, i; |
429 | ||
430 | done = 0; | |
8d1cc86a | 431 | |
bea3348e SH |
432 | poll_more: |
433 | while (done < budget) { | |
434 | int max = (budget - done); | |
435 | ||
8d1cc86a | 436 | t = min(IPOIB_NUM_WC, max); |
f56bcd80 | 437 | n = ib_poll_cq(priv->recv_cq, t, priv->ibwc); |
8d1cc86a | 438 | |
bea3348e | 439 | for (i = 0; i < n; i++) { |
8d1cc86a RD |
440 | struct ib_wc *wc = priv->ibwc + i; |
441 | ||
1b524963 | 442 | if (wc->wr_id & IPOIB_OP_RECV) { |
8d1cc86a | 443 | ++done; |
1b524963 MT |
444 | if (wc->wr_id & IPOIB_OP_CM) |
445 | ipoib_cm_handle_rx_wc(dev, wc); | |
446 | else | |
447 | ipoib_ib_handle_rx_wc(dev, wc); | |
f56bcd80 EC |
448 | } else |
449 | ipoib_cm_handle_tx_wc(priv->dev, wc); | |
8d1cc86a RD |
450 | } |
451 | ||
bea3348e | 452 | if (n != t) |
8d1cc86a | 453 | break; |
8d1cc86a RD |
454 | } |
455 | ||
bea3348e | 456 | if (done < budget) { |
288379f0 | 457 | napi_complete(napi); |
f56bcd80 | 458 | if (unlikely(ib_req_notify_cq(priv->recv_cq, |
8d1cc86a RD |
459 | IB_CQ_NEXT_COMP | |
460 | IB_CQ_REPORT_MISSED_EVENTS)) && | |
288379f0 | 461 | napi_reschedule(napi)) |
bea3348e | 462 | goto poll_more; |
8d1cc86a RD |
463 | } |
464 | ||
bea3348e | 465 | return done; |
1da177e4 LT |
466 | } |
467 | ||
468 | void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr) | |
469 | { | |
bea3348e SH |
470 | struct net_device *dev = dev_ptr; |
471 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
472 | ||
288379f0 | 473 | napi_schedule(&priv->napi); |
1da177e4 LT |
474 | } |
475 | ||
57ce41d1 EC |
476 | static void drain_tx_cq(struct net_device *dev) |
477 | { | |
478 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
57ce41d1 | 479 | |
943c246e | 480 | netif_tx_lock(dev); |
57ce41d1 EC |
481 | while (poll_tx(priv)) |
482 | ; /* nothing */ | |
483 | ||
484 | if (netif_queue_stopped(dev)) | |
485 | mod_timer(&priv->poll_timer, jiffies + 1); | |
486 | ||
943c246e | 487 | netif_tx_unlock(dev); |
57ce41d1 EC |
488 | } |
489 | ||
490 | void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr) | |
491 | { | |
943c246e RD |
492 | struct ipoib_dev_priv *priv = netdev_priv(dev_ptr); |
493 | ||
494 | mod_timer(&priv->poll_timer, jiffies); | |
57ce41d1 EC |
495 | } |
496 | ||
1da177e4 LT |
497 | static inline int post_send(struct ipoib_dev_priv *priv, |
498 | unsigned int wr_id, | |
499 | struct ib_ah *address, u32 qpn, | |
40ca1988 EC |
500 | struct ipoib_tx_buf *tx_req, |
501 | void *head, int hlen) | |
1da177e4 LT |
502 | { |
503 | struct ib_send_wr *bad_wr; | |
40ca1988 EC |
504 | int i, off; |
505 | struct sk_buff *skb = tx_req->skb; | |
506 | skb_frag_t *frags = skb_shinfo(skb)->frags; | |
507 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
508 | u64 *mapping = tx_req->mapping; | |
509 | ||
510 | if (skb_headlen(skb)) { | |
511 | priv->tx_sge[0].addr = mapping[0]; | |
512 | priv->tx_sge[0].length = skb_headlen(skb); | |
513 | off = 1; | |
514 | } else | |
515 | off = 0; | |
1da177e4 | 516 | |
7143740d | 517 | for (i = 0; i < nr_frags; ++i) { |
40ca1988 | 518 | priv->tx_sge[i + off].addr = mapping[i + off]; |
9e903e08 | 519 | priv->tx_sge[i + off].length = skb_frag_size(&frags[i]); |
7143740d | 520 | } |
40ca1988 | 521 | priv->tx_wr.num_sge = nr_frags + off; |
7143740d EC |
522 | priv->tx_wr.wr_id = wr_id; |
523 | priv->tx_wr.wr.ud.remote_qpn = qpn; | |
524 | priv->tx_wr.wr.ud.ah = address; | |
1da177e4 | 525 | |
40ca1988 EC |
526 | if (head) { |
527 | priv->tx_wr.wr.ud.mss = skb_shinfo(skb)->gso_size; | |
528 | priv->tx_wr.wr.ud.header = head; | |
529 | priv->tx_wr.wr.ud.hlen = hlen; | |
530 | priv->tx_wr.opcode = IB_WR_LSO; | |
531 | } else | |
532 | priv->tx_wr.opcode = IB_WR_SEND; | |
533 | ||
1da177e4 LT |
534 | return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr); |
535 | } | |
536 | ||
537 | void ipoib_send(struct net_device *dev, struct sk_buff *skb, | |
538 | struct ipoib_ah *address, u32 qpn) | |
539 | { | |
540 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1993d683 | 541 | struct ipoib_tx_buf *tx_req; |
a48f509b | 542 | int hlen, rc; |
40ca1988 EC |
543 | void *phead; |
544 | ||
545 | if (skb_is_gso(skb)) { | |
546 | hlen = skb_transport_offset(skb) + tcp_hdrlen(skb); | |
547 | phead = skb->data; | |
548 | if (unlikely(!skb_pull(skb, hlen))) { | |
549 | ipoib_warn(priv, "linear data too small\n"); | |
550 | ++dev->stats.tx_dropped; | |
551 | ++dev->stats.tx_errors; | |
552 | dev_kfree_skb_any(skb); | |
553 | return; | |
554 | } | |
555 | } else { | |
556 | if (unlikely(skb->len > priv->mcast_mtu + IPOIB_ENCAP_LEN)) { | |
557 | ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n", | |
558 | skb->len, priv->mcast_mtu + IPOIB_ENCAP_LEN); | |
559 | ++dev->stats.tx_dropped; | |
560 | ++dev->stats.tx_errors; | |
561 | ipoib_cm_skb_too_long(dev, skb, priv->mcast_mtu); | |
562 | return; | |
563 | } | |
564 | phead = NULL; | |
565 | hlen = 0; | |
1da177e4 LT |
566 | } |
567 | ||
568 | ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n", | |
569 | skb->len, address, qpn); | |
570 | ||
571 | /* | |
572 | * We put the skb into the tx_ring _before_ we call post_send() | |
573 | * because it's entirely possible that the completion handler will | |
574 | * run before we execute anything after the post_send(). That | |
575 | * means we have to make sure everything is properly recorded and | |
576 | * our state is consistent before we call post_send(). | |
577 | */ | |
0f485251 | 578 | tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)]; |
1da177e4 | 579 | tx_req->skb = skb; |
7143740d | 580 | if (unlikely(ipoib_dma_map_tx(priv->ca, tx_req))) { |
de903512 | 581 | ++dev->stats.tx_errors; |
73fbe8be RD |
582 | dev_kfree_skb_any(skb); |
583 | return; | |
584 | } | |
1da177e4 | 585 | |
6046136c EC |
586 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
587 | priv->tx_wr.send_flags |= IB_SEND_IP_CSUM; | |
588 | else | |
589 | priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM; | |
590 | ||
57ce41d1 EC |
591 | if (++priv->tx_outstanding == ipoib_sendq_size) { |
592 | ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n"); | |
593 | if (ib_req_notify_cq(priv->send_cq, IB_CQ_NEXT_COMP)) | |
594 | ipoib_warn(priv, "request notify on send CQ failed\n"); | |
595 | netif_stop_queue(dev); | |
596 | } | |
597 | ||
a48f509b OG |
598 | rc = post_send(priv, priv->tx_head & (ipoib_sendq_size - 1), |
599 | address->ah, qpn, tx_req, phead, hlen); | |
600 | if (unlikely(rc)) { | |
601 | ipoib_warn(priv, "post_send failed, error %d\n", rc); | |
de903512 | 602 | ++dev->stats.tx_errors; |
57ce41d1 | 603 | --priv->tx_outstanding; |
7143740d | 604 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
1da177e4 | 605 | dev_kfree_skb_any(skb); |
57ce41d1 EC |
606 | if (netif_queue_stopped(dev)) |
607 | netif_wake_queue(dev); | |
1da177e4 LT |
608 | } else { |
609 | dev->trans_start = jiffies; | |
610 | ||
611 | address->last_send = priv->tx_head; | |
612 | ++priv->tx_head; | |
f56bcd80 | 613 | skb_orphan(skb); |
1da177e4 | 614 | |
1da177e4 | 615 | } |
f56bcd80 EC |
616 | |
617 | if (unlikely(priv->tx_outstanding > MAX_SEND_CQE)) | |
57ce41d1 EC |
618 | while (poll_tx(priv)) |
619 | ; /* nothing */ | |
1da177e4 LT |
620 | } |
621 | ||
622 | static void __ipoib_reap_ah(struct net_device *dev) | |
623 | { | |
624 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
625 | struct ipoib_ah *ah, *tah; | |
626 | LIST_HEAD(remove_list); | |
943c246e RD |
627 | unsigned long flags; |
628 | ||
629 | netif_tx_lock_bh(dev); | |
630 | spin_lock_irqsave(&priv->lock, flags); | |
1da177e4 | 631 | |
1da177e4 | 632 | list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list) |
2181858b | 633 | if ((int) priv->tx_tail - (int) ah->last_send >= 0) { |
1da177e4 | 634 | list_del(&ah->list); |
31c02e21 RD |
635 | ib_destroy_ah(ah->ah); |
636 | kfree(ah); | |
1da177e4 | 637 | } |
943c246e RD |
638 | |
639 | spin_unlock_irqrestore(&priv->lock, flags); | |
640 | netif_tx_unlock_bh(dev); | |
1da177e4 LT |
641 | } |
642 | ||
c4028958 | 643 | void ipoib_reap_ah(struct work_struct *work) |
1da177e4 | 644 | { |
c4028958 DH |
645 | struct ipoib_dev_priv *priv = |
646 | container_of(work, struct ipoib_dev_priv, ah_reap_task.work); | |
647 | struct net_device *dev = priv->dev; | |
1da177e4 LT |
648 | |
649 | __ipoib_reap_ah(dev); | |
650 | ||
651 | if (!test_bit(IPOIB_STOP_REAPER, &priv->flags)) | |
69fc507a AB |
652 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, |
653 | round_jiffies_relative(HZ)); | |
1da177e4 LT |
654 | } |
655 | ||
57ce41d1 EC |
656 | static void ipoib_ib_tx_timer_func(unsigned long ctx) |
657 | { | |
658 | drain_tx_cq((struct net_device *)ctx); | |
659 | } | |
660 | ||
1da177e4 LT |
661 | int ipoib_ib_dev_open(struct net_device *dev) |
662 | { | |
663 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
664 | int ret; | |
665 | ||
26bbf13c YE |
666 | if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &priv->pkey_index)) { |
667 | ipoib_warn(priv, "P_Key 0x%04x not found\n", priv->pkey); | |
668 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
669 | return -1; | |
670 | } | |
671 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
672 | ||
5b6810e0 | 673 | ret = ipoib_init_qp(dev); |
1da177e4 | 674 | if (ret) { |
5b6810e0 | 675 | ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret); |
1da177e4 LT |
676 | return -1; |
677 | } | |
678 | ||
679 | ret = ipoib_ib_post_receives(dev); | |
680 | if (ret) { | |
681 | ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret); | |
26bbf13c | 682 | ipoib_ib_dev_stop(dev, 1); |
1da177e4 LT |
683 | return -1; |
684 | } | |
685 | ||
839fcaba MT |
686 | ret = ipoib_cm_dev_open(dev); |
687 | if (ret) { | |
24bd1e4e | 688 | ipoib_warn(priv, "ipoib_cm_dev_open returned %d\n", ret); |
26bbf13c | 689 | ipoib_ib_dev_stop(dev, 1); |
839fcaba MT |
690 | return -1; |
691 | } | |
692 | ||
1da177e4 | 693 | clear_bit(IPOIB_STOP_REAPER, &priv->flags); |
69fc507a AB |
694 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, |
695 | round_jiffies_relative(HZ)); | |
1da177e4 | 696 | |
e028cc55 YE |
697 | if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
698 | napi_enable(&priv->napi); | |
7a343d4c | 699 | |
1da177e4 LT |
700 | return 0; |
701 | } | |
702 | ||
7a343d4c LA |
703 | static void ipoib_pkey_dev_check_presence(struct net_device *dev) |
704 | { | |
705 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
706 | u16 pkey_index = 0; | |
707 | ||
9fdd5e5b | 708 | if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) |
7a343d4c LA |
709 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); |
710 | else | |
711 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
712 | } | |
713 | ||
1da177e4 LT |
714 | int ipoib_ib_dev_up(struct net_device *dev) |
715 | { | |
716 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
717 | ||
7a343d4c LA |
718 | ipoib_pkey_dev_check_presence(dev); |
719 | ||
720 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
721 | ipoib_dbg(priv, "PKEY is not assigned.\n"); | |
722 | return 0; | |
723 | } | |
724 | ||
1da177e4 LT |
725 | set_bit(IPOIB_FLAG_OPER_UP, &priv->flags); |
726 | ||
727 | return ipoib_mcast_start_thread(dev); | |
728 | } | |
729 | ||
0b3ea082 | 730 | int ipoib_ib_dev_down(struct net_device *dev, int flush) |
1da177e4 LT |
731 | { |
732 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
733 | ||
734 | ipoib_dbg(priv, "downing ib_dev\n"); | |
735 | ||
736 | clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags); | |
737 | netif_carrier_off(dev); | |
738 | ||
739 | /* Shutdown the P_Key thread if still active */ | |
740 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
95ed644f | 741 | mutex_lock(&pkey_mutex); |
1da177e4 | 742 | set_bit(IPOIB_PKEY_STOP, &priv->flags); |
26bbf13c | 743 | cancel_delayed_work(&priv->pkey_poll_task); |
95ed644f | 744 | mutex_unlock(&pkey_mutex); |
0b3ea082 JM |
745 | if (flush) |
746 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
747 | } |
748 | ||
0b3ea082 | 749 | ipoib_mcast_stop_thread(dev, flush); |
1da177e4 LT |
750 | ipoib_mcast_dev_flush(dev); |
751 | ||
1da177e4 LT |
752 | ipoib_flush_paths(dev); |
753 | ||
754 | return 0; | |
755 | } | |
756 | ||
757 | static int recvs_pending(struct net_device *dev) | |
758 | { | |
759 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
760 | int pending = 0; | |
761 | int i; | |
762 | ||
0f485251 | 763 | for (i = 0; i < ipoib_recvq_size; ++i) |
1da177e4 LT |
764 | if (priv->rx_ring[i].skb) |
765 | ++pending; | |
766 | ||
767 | return pending; | |
768 | } | |
769 | ||
2dfbfc37 MT |
770 | void ipoib_drain_cq(struct net_device *dev) |
771 | { | |
772 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
773 | int i, n; | |
943c246e RD |
774 | |
775 | /* | |
776 | * We call completion handling routines that expect to be | |
777 | * called from the BH-disabled NAPI poll context, so disable | |
778 | * BHs here too. | |
779 | */ | |
780 | local_bh_disable(); | |
781 | ||
2dfbfc37 | 782 | do { |
f56bcd80 | 783 | n = ib_poll_cq(priv->recv_cq, IPOIB_NUM_WC, priv->ibwc); |
2dfbfc37 | 784 | for (i = 0; i < n; ++i) { |
ce423ef5 RD |
785 | /* |
786 | * Convert any successful completions to flush | |
787 | * errors to avoid passing packets up the | |
788 | * stack after bringing the device down. | |
789 | */ | |
790 | if (priv->ibwc[i].status == IB_WC_SUCCESS) | |
791 | priv->ibwc[i].status = IB_WC_WR_FLUSH_ERR; | |
792 | ||
1b524963 MT |
793 | if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) { |
794 | if (priv->ibwc[i].wr_id & IPOIB_OP_CM) | |
795 | ipoib_cm_handle_rx_wc(dev, priv->ibwc + i); | |
796 | else | |
797 | ipoib_ib_handle_rx_wc(dev, priv->ibwc + i); | |
f56bcd80 EC |
798 | } else |
799 | ipoib_cm_handle_tx_wc(dev, priv->ibwc + i); | |
2dfbfc37 MT |
800 | } |
801 | } while (n == IPOIB_NUM_WC); | |
f56bcd80 EC |
802 | |
803 | while (poll_tx(priv)) | |
804 | ; /* nothing */ | |
943c246e RD |
805 | |
806 | local_bh_enable(); | |
2dfbfc37 MT |
807 | } |
808 | ||
26bbf13c | 809 | int ipoib_ib_dev_stop(struct net_device *dev, int flush) |
1da177e4 LT |
810 | { |
811 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
812 | struct ib_qp_attr qp_attr; | |
1da177e4 | 813 | unsigned long begin; |
1993d683 | 814 | struct ipoib_tx_buf *tx_req; |
2dfbfc37 | 815 | int i; |
1da177e4 | 816 | |
e028cc55 YE |
817 | if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
818 | napi_disable(&priv->napi); | |
7a343d4c | 819 | |
839fcaba MT |
820 | ipoib_cm_dev_stop(dev); |
821 | ||
3bc12e75 RD |
822 | /* |
823 | * Move our QP to the error state and then reinitialize in | |
824 | * when all work requests have completed or have been flushed. | |
825 | */ | |
1da177e4 | 826 | qp_attr.qp_state = IB_QPS_ERR; |
3bc12e75 | 827 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
828 | ipoib_warn(priv, "Failed to modify QP to ERROR state\n"); |
829 | ||
830 | /* Wait for all sends and receives to complete */ | |
831 | begin = jiffies; | |
832 | ||
833 | while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) { | |
834 | if (time_after(jiffies, begin + 5 * HZ)) { | |
835 | ipoib_warn(priv, "timing out; %d sends %d receives not completed\n", | |
836 | priv->tx_head - priv->tx_tail, recvs_pending(dev)); | |
837 | ||
838 | /* | |
839 | * assume the HW is wedged and just free up | |
840 | * all our pending work requests. | |
841 | */ | |
2181858b | 842 | while ((int) priv->tx_tail - (int) priv->tx_head < 0) { |
1da177e4 | 843 | tx_req = &priv->tx_ring[priv->tx_tail & |
0f485251 | 844 | (ipoib_sendq_size - 1)]; |
7143740d | 845 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
1da177e4 LT |
846 | dev_kfree_skb_any(tx_req->skb); |
847 | ++priv->tx_tail; | |
1b524963 | 848 | --priv->tx_outstanding; |
1da177e4 LT |
849 | } |
850 | ||
37ccf9df RC |
851 | for (i = 0; i < ipoib_recvq_size; ++i) { |
852 | struct ipoib_rx_buf *rx_req; | |
853 | ||
854 | rx_req = &priv->rx_ring[i]; | |
855 | if (!rx_req->skb) | |
856 | continue; | |
bc7b3a36 SM |
857 | ipoib_ud_dma_unmap_rx(priv, |
858 | priv->rx_ring[i].mapping); | |
37ccf9df RC |
859 | dev_kfree_skb_any(rx_req->skb); |
860 | rx_req->skb = NULL; | |
861 | } | |
1da177e4 LT |
862 | |
863 | goto timeout; | |
864 | } | |
865 | ||
2dfbfc37 | 866 | ipoib_drain_cq(dev); |
8d1cc86a | 867 | |
1da177e4 LT |
868 | msleep(1); |
869 | } | |
870 | ||
871 | ipoib_dbg(priv, "All sends and receives done.\n"); | |
872 | ||
873 | timeout: | |
57ce41d1 | 874 | del_timer_sync(&priv->poll_timer); |
1da177e4 | 875 | qp_attr.qp_state = IB_QPS_RESET; |
3bc12e75 | 876 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
877 | ipoib_warn(priv, "Failed to modify QP to RESET state\n"); |
878 | ||
879 | /* Wait for all AHs to be reaped */ | |
880 | set_bit(IPOIB_STOP_REAPER, &priv->flags); | |
881 | cancel_delayed_work(&priv->ah_reap_task); | |
26bbf13c YE |
882 | if (flush) |
883 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
884 | |
885 | begin = jiffies; | |
886 | ||
887 | while (!list_empty(&priv->dead_ahs)) { | |
888 | __ipoib_reap_ah(dev); | |
889 | ||
890 | if (time_after(jiffies, begin + HZ)) { | |
891 | ipoib_warn(priv, "timing out; will leak address handles\n"); | |
892 | break; | |
893 | } | |
894 | ||
895 | msleep(1); | |
896 | } | |
897 | ||
f56bcd80 | 898 | ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP); |
8d1cc86a | 899 | |
1da177e4 LT |
900 | return 0; |
901 | } | |
902 | ||
903 | int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port) | |
904 | { | |
905 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
906 | ||
907 | priv->ca = ca; | |
908 | priv->port = port; | |
909 | priv->qp = NULL; | |
910 | ||
911 | if (ipoib_transport_dev_init(dev, ca)) { | |
912 | printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name); | |
913 | return -ENODEV; | |
914 | } | |
915 | ||
2767840a RD |
916 | setup_timer(&priv->poll_timer, ipoib_ib_tx_timer_func, |
917 | (unsigned long) dev); | |
918 | ||
1da177e4 LT |
919 | if (dev->flags & IFF_UP) { |
920 | if (ipoib_ib_dev_open(dev)) { | |
921 | ipoib_transport_dev_cleanup(dev); | |
922 | return -ENODEV; | |
923 | } | |
924 | } | |
925 | ||
926 | return 0; | |
927 | } | |
928 | ||
ee1e2c82 MS |
929 | static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, |
930 | enum ipoib_flush_level level) | |
1da177e4 | 931 | { |
26bbf13c | 932 | struct ipoib_dev_priv *cpriv; |
c4028958 | 933 | struct net_device *dev = priv->dev; |
26bbf13c YE |
934 | u16 new_index; |
935 | ||
936 | mutex_lock(&priv->vlan_mutex); | |
1da177e4 | 937 | |
26bbf13c YE |
938 | /* |
939 | * Flush any child interfaces too -- they might be up even if | |
940 | * the parent is down. | |
941 | */ | |
942 | list_for_each_entry(cpriv, &priv->child_intfs, list) | |
ee1e2c82 | 943 | __ipoib_ib_dev_flush(cpriv, level); |
26bbf13c YE |
944 | |
945 | mutex_unlock(&priv->vlan_mutex); | |
946 | ||
947 | if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) { | |
7a343d4c LA |
948 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n"); |
949 | return; | |
950 | } | |
951 | ||
952 | if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { | |
953 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n"); | |
1da177e4 | 954 | return; |
7a343d4c | 955 | } |
1da177e4 | 956 | |
ee1e2c82 | 957 | if (level == IPOIB_FLUSH_HEAVY) { |
26bbf13c YE |
958 | if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &new_index)) { |
959 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
960 | ipoib_ib_dev_down(dev, 0); | |
167c4265 | 961 | ipoib_ib_dev_stop(dev, 0); |
9fdd5e5b RD |
962 | if (ipoib_pkey_dev_delay_open(dev)) |
963 | return; | |
26bbf13c | 964 | } |
26bbf13c YE |
965 | |
966 | /* restart QP only if P_Key index is changed */ | |
9fdd5e5b RD |
967 | if (test_and_set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags) && |
968 | new_index == priv->pkey_index) { | |
26bbf13c YE |
969 | ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n"); |
970 | return; | |
971 | } | |
972 | priv->pkey_index = new_index; | |
973 | } | |
974 | ||
ee1e2c82 MS |
975 | if (level == IPOIB_FLUSH_LIGHT) { |
976 | ipoib_mark_paths_invalid(dev); | |
977 | ipoib_mcast_dev_flush(dev); | |
978 | } | |
1da177e4 | 979 | |
ee1e2c82 MS |
980 | if (level >= IPOIB_FLUSH_NORMAL) |
981 | ipoib_ib_dev_down(dev, 0); | |
1da177e4 | 982 | |
ee1e2c82 | 983 | if (level == IPOIB_FLUSH_HEAVY) { |
26bbf13c YE |
984 | ipoib_ib_dev_stop(dev, 0); |
985 | ipoib_ib_dev_open(dev); | |
986 | } | |
987 | ||
1da177e4 LT |
988 | /* |
989 | * The device could have been brought down between the start and when | |
990 | * we get here, don't bring it back up if it's not configured up | |
991 | */ | |
5ccd0255 | 992 | if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { |
ee1e2c82 MS |
993 | if (level >= IPOIB_FLUSH_NORMAL) |
994 | ipoib_ib_dev_up(dev); | |
c4028958 | 995 | ipoib_mcast_restart_task(&priv->restart_task); |
5ccd0255 | 996 | } |
26bbf13c | 997 | } |
1da177e4 | 998 | |
ee1e2c82 MS |
999 | void ipoib_ib_dev_flush_light(struct work_struct *work) |
1000 | { | |
1001 | struct ipoib_dev_priv *priv = | |
1002 | container_of(work, struct ipoib_dev_priv, flush_light); | |
1003 | ||
1004 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_LIGHT); | |
1005 | } | |
1006 | ||
1007 | void ipoib_ib_dev_flush_normal(struct work_struct *work) | |
26bbf13c YE |
1008 | { |
1009 | struct ipoib_dev_priv *priv = | |
ee1e2c82 | 1010 | container_of(work, struct ipoib_dev_priv, flush_normal); |
4f71055a | 1011 | |
ee1e2c82 | 1012 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_NORMAL); |
26bbf13c | 1013 | } |
4f71055a | 1014 | |
ee1e2c82 | 1015 | void ipoib_ib_dev_flush_heavy(struct work_struct *work) |
26bbf13c YE |
1016 | { |
1017 | struct ipoib_dev_priv *priv = | |
ee1e2c82 | 1018 | container_of(work, struct ipoib_dev_priv, flush_heavy); |
26bbf13c | 1019 | |
ee1e2c82 | 1020 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_HEAVY); |
1da177e4 LT |
1021 | } |
1022 | ||
1023 | void ipoib_ib_dev_cleanup(struct net_device *dev) | |
1024 | { | |
1025 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1026 | ||
1027 | ipoib_dbg(priv, "cleaning up ib_dev\n"); | |
1028 | ||
8d2cae06 | 1029 | ipoib_mcast_stop_thread(dev, 1); |
988bd503 | 1030 | ipoib_mcast_dev_flush(dev); |
1da177e4 LT |
1031 | |
1032 | ipoib_transport_dev_cleanup(dev); | |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * Delayed P_Key Assigment Interim Support | |
1037 | * | |
1038 | * The following is initial implementation of delayed P_Key assigment | |
1039 | * mechanism. It is using the same approach implemented for the multicast | |
1040 | * group join. The single goal of this implementation is to quickly address | |
1041 | * Bug #2507. This implementation will probably be removed when the P_Key | |
1042 | * change async notification is available. | |
1043 | */ | |
1da177e4 | 1044 | |
c4028958 | 1045 | void ipoib_pkey_poll(struct work_struct *work) |
1da177e4 | 1046 | { |
c4028958 | 1047 | struct ipoib_dev_priv *priv = |
26bbf13c | 1048 | container_of(work, struct ipoib_dev_priv, pkey_poll_task.work); |
c4028958 | 1049 | struct net_device *dev = priv->dev; |
1da177e4 LT |
1050 | |
1051 | ipoib_pkey_dev_check_presence(dev); | |
1052 | ||
1053 | if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) | |
1054 | ipoib_open(dev); | |
1055 | else { | |
95ed644f | 1056 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
1057 | if (!test_bit(IPOIB_PKEY_STOP, &priv->flags)) |
1058 | queue_delayed_work(ipoib_workqueue, | |
26bbf13c | 1059 | &priv->pkey_poll_task, |
1da177e4 | 1060 | HZ); |
95ed644f | 1061 | mutex_unlock(&pkey_mutex); |
1da177e4 LT |
1062 | } |
1063 | } | |
1064 | ||
1065 | int ipoib_pkey_dev_delay_open(struct net_device *dev) | |
1066 | { | |
1067 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1068 | ||
1069 | /* Look for the interface pkey value in the IB Port P_Key table and */ | |
1070 | /* set the interface pkey assigment flag */ | |
1071 | ipoib_pkey_dev_check_presence(dev); | |
1072 | ||
1073 | /* P_Key value not assigned yet - start polling */ | |
1074 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
95ed644f | 1075 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
1076 | clear_bit(IPOIB_PKEY_STOP, &priv->flags); |
1077 | queue_delayed_work(ipoib_workqueue, | |
26bbf13c | 1078 | &priv->pkey_poll_task, |
1da177e4 | 1079 | HZ); |
95ed644f | 1080 | mutex_unlock(&pkey_mutex); |
1da177e4 LT |
1081 | return 1; |
1082 | } | |
1083 | ||
1084 | return 0; | |
1085 | } |