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 | 125 | skb->data_len += size; |
b28ba726 | 126 | skb->truesize += PAGE_SIZE; |
bc7b3a36 SM |
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 | 158 | int buf_size; |
b28ba726 | 159 | int tailroom; |
bc7b3a36 | 160 | u64 *mapping; |
1da177e4 | 161 | |
b28ba726 | 162 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) { |
bc7b3a36 | 163 | buf_size = IPOIB_UD_HEAD_SIZE; |
b28ba726 ED |
164 | tailroom = 128; /* reserve some tailroom for IP/TCP headers */ |
165 | } else { | |
bc7b3a36 | 166 | buf_size = IPOIB_UD_BUF_SIZE(priv->max_ib_mtu); |
b28ba726 ED |
167 | tailroom = 0; |
168 | } | |
bc7b3a36 | 169 | |
b28ba726 | 170 | skb = dev_alloc_skb(buf_size + tailroom + 4); |
bc7b3a36 SM |
171 | if (unlikely(!skb)) |
172 | return NULL; | |
1993d683 RD |
173 | |
174 | /* | |
175 | * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte | |
176 | * header. So we need 4 more bytes to get to 48 and align the | |
177 | * IP header to a multiple of 16. | |
178 | */ | |
179 | skb_reserve(skb, 4); | |
180 | ||
bc7b3a36 SM |
181 | mapping = priv->rx_ring[id].mapping; |
182 | mapping[0] = ib_dma_map_single(priv->ca, skb->data, buf_size, | |
183 | DMA_FROM_DEVICE); | |
184 | if (unlikely(ib_dma_mapping_error(priv->ca, mapping[0]))) | |
185 | goto error; | |
186 | ||
187 | if (ipoib_ud_need_sg(priv->max_ib_mtu)) { | |
188 | struct page *page = alloc_page(GFP_ATOMIC); | |
189 | if (!page) | |
190 | goto partial_error; | |
191 | skb_fill_page_desc(skb, 0, page, 0, PAGE_SIZE); | |
192 | mapping[1] = | |
5581be3b | 193 | ib_dma_map_page(priv->ca, page, |
bc7b3a36 SM |
194 | 0, PAGE_SIZE, DMA_FROM_DEVICE); |
195 | if (unlikely(ib_dma_mapping_error(priv->ca, mapping[1]))) | |
196 | goto partial_error; | |
1da177e4 LT |
197 | } |
198 | ||
bc7b3a36 SM |
199 | priv->rx_ring[id].skb = skb; |
200 | return skb; | |
1993d683 | 201 | |
bc7b3a36 SM |
202 | partial_error: |
203 | ib_dma_unmap_single(priv->ca, mapping[0], buf_size, DMA_FROM_DEVICE); | |
204 | error: | |
205 | dev_kfree_skb_any(skb); | |
206 | return NULL; | |
1da177e4 LT |
207 | } |
208 | ||
209 | static int ipoib_ib_post_receives(struct net_device *dev) | |
210 | { | |
211 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
212 | int i; | |
213 | ||
0f485251 | 214 | for (i = 0; i < ipoib_recvq_size; ++i) { |
bc7b3a36 | 215 | if (!ipoib_alloc_rx_skb(dev, i)) { |
1993d683 RD |
216 | ipoib_warn(priv, "failed to allocate receive buffer %d\n", i); |
217 | return -ENOMEM; | |
218 | } | |
1da177e4 LT |
219 | if (ipoib_ib_post_receive(dev, i)) { |
220 | ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i); | |
221 | return -EIO; | |
222 | } | |
223 | } | |
224 | ||
225 | return 0; | |
226 | } | |
227 | ||
2439a6e6 | 228 | static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) |
1da177e4 LT |
229 | { |
230 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
2439a6e6 RD |
231 | unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV; |
232 | struct sk_buff *skb; | |
bc7b3a36 | 233 | u64 mapping[IPOIB_UD_RX_SG]; |
fed1db33 | 234 | union ib_gid *dgid; |
1da177e4 | 235 | |
a89875fc RD |
236 | ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n", |
237 | wr_id, wc->status); | |
1da177e4 | 238 | |
2439a6e6 RD |
239 | if (unlikely(wr_id >= ipoib_recvq_size)) { |
240 | ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n", | |
241 | wr_id, ipoib_recvq_size); | |
242 | return; | |
243 | } | |
244 | ||
245 | skb = priv->rx_ring[wr_id].skb; | |
2439a6e6 RD |
246 | |
247 | if (unlikely(wc->status != IB_WC_SUCCESS)) { | |
248 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
249 | ipoib_warn(priv, "failed recv event " | |
250 | "(status=%d, wrid=%d vend_err %x)\n", | |
251 | wc->status, wr_id, wc->vendor_err); | |
bc7b3a36 | 252 | ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[wr_id].mapping); |
2439a6e6 RD |
253 | dev_kfree_skb_any(skb); |
254 | priv->rx_ring[wr_id].skb = NULL; | |
255 | return; | |
256 | } | |
1da177e4 | 257 | |
1b844afe RD |
258 | /* |
259 | * Drop packets that this interface sent, ie multicast packets | |
260 | * that the HCA has replicated. | |
261 | */ | |
262 | if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num) | |
263 | goto repost; | |
264 | ||
bc7b3a36 SM |
265 | memcpy(mapping, priv->rx_ring[wr_id].mapping, |
266 | IPOIB_UD_RX_SG * sizeof *mapping); | |
267 | ||
2439a6e6 RD |
268 | /* |
269 | * If we can't allocate a new RX buffer, dump | |
270 | * this packet and reuse the old buffer. | |
271 | */ | |
bc7b3a36 | 272 | if (unlikely(!ipoib_alloc_rx_skb(dev, wr_id))) { |
de903512 | 273 | ++dev->stats.rx_dropped; |
2439a6e6 RD |
274 | goto repost; |
275 | } | |
276 | ||
277 | ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n", | |
278 | wc->byte_len, wc->slid); | |
279 | ||
bc7b3a36 SM |
280 | ipoib_ud_dma_unmap_rx(priv, mapping); |
281 | ipoib_ud_skb_put_frags(priv, skb, wc->byte_len); | |
2439a6e6 | 282 | |
fed1db33 CL |
283 | /* First byte of dgid signals multicast when 0xff */ |
284 | dgid = &((struct ib_grh *)skb->data)->dgid; | |
285 | ||
286 | if (!(wc->wc_flags & IB_WC_GRH) || dgid->raw[0] != 0xff) | |
287 | skb->pkt_type = PACKET_HOST; | |
288 | else if (memcmp(dgid, dev->broadcast + 4, sizeof(union ib_gid)) == 0) | |
289 | skb->pkt_type = PACKET_BROADCAST; | |
290 | else | |
291 | skb->pkt_type = PACKET_MULTICAST; | |
292 | ||
2439a6e6 RD |
293 | skb_pull(skb, IB_GRH_BYTES); |
294 | ||
1b844afe RD |
295 | skb->protocol = ((struct ipoib_header *) skb->data)->proto; |
296 | skb_reset_mac_header(skb); | |
297 | skb_pull(skb, IPOIB_ENCAP_LEN); | |
298 | ||
de903512 RD |
299 | ++dev->stats.rx_packets; |
300 | dev->stats.rx_bytes += skb->len; | |
1b844afe RD |
301 | |
302 | skb->dev = dev; | |
d927d505 OG |
303 | if ((dev->features & NETIF_F_RXCSUM) && |
304 | likely(wc->wc_flags & IB_WC_IP_CSUM_OK)) | |
6046136c EC |
305 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
306 | ||
8ae31e5b | 307 | napi_gro_receive(&priv->napi, skb); |
1da177e4 | 308 | |
2439a6e6 RD |
309 | repost: |
310 | if (unlikely(ipoib_ib_post_receive(dev, wr_id))) | |
311 | ipoib_warn(priv, "ipoib_ib_post_receive failed " | |
312 | "for buf %d\n", wr_id); | |
313 | } | |
1da177e4 | 314 | |
7143740d EC |
315 | static int ipoib_dma_map_tx(struct ib_device *ca, |
316 | struct ipoib_tx_buf *tx_req) | |
317 | { | |
318 | struct sk_buff *skb = tx_req->skb; | |
319 | u64 *mapping = tx_req->mapping; | |
320 | int i; | |
40ca1988 | 321 | int off; |
7143740d | 322 | |
40ca1988 EC |
323 | if (skb_headlen(skb)) { |
324 | mapping[0] = ib_dma_map_single(ca, skb->data, skb_headlen(skb), | |
325 | DMA_TO_DEVICE); | |
326 | if (unlikely(ib_dma_mapping_error(ca, mapping[0]))) | |
327 | return -EIO; | |
328 | ||
329 | off = 1; | |
330 | } else | |
331 | off = 0; | |
7143740d EC |
332 | |
333 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
9e903e08 | 334 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
5581be3b IC |
335 | mapping[i + off] = ib_dma_map_page(ca, |
336 | skb_frag_page(frag), | |
9e903e08 | 337 | frag->page_offset, skb_frag_size(frag), |
7143740d | 338 | DMA_TO_DEVICE); |
40ca1988 | 339 | if (unlikely(ib_dma_mapping_error(ca, mapping[i + off]))) |
7143740d EC |
340 | goto partial_error; |
341 | } | |
342 | return 0; | |
343 | ||
344 | partial_error: | |
7143740d | 345 | for (; i > 0; --i) { |
9e903e08 ED |
346 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; |
347 | ||
348 | ib_dma_unmap_page(ca, mapping[i - !off], skb_frag_size(frag), DMA_TO_DEVICE); | |
7143740d | 349 | } |
40ca1988 EC |
350 | |
351 | if (off) | |
352 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
353 | ||
7143740d EC |
354 | return -EIO; |
355 | } | |
356 | ||
357 | static void ipoib_dma_unmap_tx(struct ib_device *ca, | |
358 | struct ipoib_tx_buf *tx_req) | |
359 | { | |
360 | struct sk_buff *skb = tx_req->skb; | |
361 | u64 *mapping = tx_req->mapping; | |
362 | int i; | |
40ca1988 | 363 | int off; |
7143740d | 364 | |
40ca1988 EC |
365 | if (skb_headlen(skb)) { |
366 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
367 | off = 1; | |
368 | } else | |
369 | off = 0; | |
7143740d EC |
370 | |
371 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
9e903e08 ED |
372 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
373 | ||
374 | ib_dma_unmap_page(ca, mapping[i + off], skb_frag_size(frag), | |
7143740d EC |
375 | DMA_TO_DEVICE); |
376 | } | |
377 | } | |
378 | ||
2439a6e6 RD |
379 | static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) |
380 | { | |
381 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
382 | unsigned int wr_id = wc->wr_id; | |
383 | struct ipoib_tx_buf *tx_req; | |
1da177e4 | 384 | |
a89875fc RD |
385 | ipoib_dbg_data(priv, "send completion: id %d, status: %d\n", |
386 | wr_id, wc->status); | |
1da177e4 | 387 | |
2439a6e6 RD |
388 | if (unlikely(wr_id >= ipoib_sendq_size)) { |
389 | ipoib_warn(priv, "send completion event with wrid %d (> %d)\n", | |
390 | wr_id, ipoib_sendq_size); | |
391 | return; | |
1da177e4 | 392 | } |
2439a6e6 RD |
393 | |
394 | tx_req = &priv->tx_ring[wr_id]; | |
395 | ||
7143740d | 396 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
2439a6e6 | 397 | |
de903512 RD |
398 | ++dev->stats.tx_packets; |
399 | dev->stats.tx_bytes += tx_req->skb->len; | |
2439a6e6 RD |
400 | |
401 | dev_kfree_skb_any(tx_req->skb); | |
402 | ||
2439a6e6 | 403 | ++priv->tx_tail; |
1b524963 MT |
404 | if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) && |
405 | netif_queue_stopped(dev) && | |
406 | test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) | |
2439a6e6 | 407 | netif_wake_queue(dev); |
2439a6e6 RD |
408 | |
409 | if (wc->status != IB_WC_SUCCESS && | |
410 | wc->status != IB_WC_WR_FLUSH_ERR) | |
411 | ipoib_warn(priv, "failed send event " | |
412 | "(status=%d, wrid=%d vend_err %x)\n", | |
413 | wc->status, wr_id, wc->vendor_err); | |
414 | } | |
415 | ||
f56bcd80 EC |
416 | static int poll_tx(struct ipoib_dev_priv *priv) |
417 | { | |
418 | int n, i; | |
419 | ||
420 | n = ib_poll_cq(priv->send_cq, MAX_SEND_CQE, priv->send_wc); | |
421 | for (i = 0; i < n; ++i) | |
422 | ipoib_ib_handle_tx_wc(priv->dev, priv->send_wc + i); | |
423 | ||
424 | return n == MAX_SEND_CQE; | |
425 | } | |
426 | ||
bea3348e | 427 | int ipoib_poll(struct napi_struct *napi, int budget) |
2439a6e6 | 428 | { |
bea3348e SH |
429 | struct ipoib_dev_priv *priv = container_of(napi, struct ipoib_dev_priv, napi); |
430 | struct net_device *dev = priv->dev; | |
8d1cc86a RD |
431 | int done; |
432 | int t; | |
8d1cc86a RD |
433 | int n, i; |
434 | ||
435 | done = 0; | |
8d1cc86a | 436 | |
bea3348e SH |
437 | poll_more: |
438 | while (done < budget) { | |
439 | int max = (budget - done); | |
440 | ||
8d1cc86a | 441 | t = min(IPOIB_NUM_WC, max); |
f56bcd80 | 442 | n = ib_poll_cq(priv->recv_cq, t, priv->ibwc); |
8d1cc86a | 443 | |
bea3348e | 444 | for (i = 0; i < n; i++) { |
8d1cc86a RD |
445 | struct ib_wc *wc = priv->ibwc + i; |
446 | ||
1b524963 | 447 | if (wc->wr_id & IPOIB_OP_RECV) { |
8d1cc86a | 448 | ++done; |
1b524963 MT |
449 | if (wc->wr_id & IPOIB_OP_CM) |
450 | ipoib_cm_handle_rx_wc(dev, wc); | |
451 | else | |
452 | ipoib_ib_handle_rx_wc(dev, wc); | |
f56bcd80 EC |
453 | } else |
454 | ipoib_cm_handle_tx_wc(priv->dev, wc); | |
8d1cc86a RD |
455 | } |
456 | ||
bea3348e | 457 | if (n != t) |
8d1cc86a | 458 | break; |
8d1cc86a RD |
459 | } |
460 | ||
bea3348e | 461 | if (done < budget) { |
288379f0 | 462 | napi_complete(napi); |
f56bcd80 | 463 | if (unlikely(ib_req_notify_cq(priv->recv_cq, |
8d1cc86a RD |
464 | IB_CQ_NEXT_COMP | |
465 | IB_CQ_REPORT_MISSED_EVENTS)) && | |
288379f0 | 466 | napi_reschedule(napi)) |
bea3348e | 467 | goto poll_more; |
8d1cc86a RD |
468 | } |
469 | ||
bea3348e | 470 | return done; |
1da177e4 LT |
471 | } |
472 | ||
473 | void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr) | |
474 | { | |
bea3348e SH |
475 | struct net_device *dev = dev_ptr; |
476 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
477 | ||
288379f0 | 478 | napi_schedule(&priv->napi); |
1da177e4 LT |
479 | } |
480 | ||
57ce41d1 EC |
481 | static void drain_tx_cq(struct net_device *dev) |
482 | { | |
483 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
57ce41d1 | 484 | |
943c246e | 485 | netif_tx_lock(dev); |
57ce41d1 EC |
486 | while (poll_tx(priv)) |
487 | ; /* nothing */ | |
488 | ||
489 | if (netif_queue_stopped(dev)) | |
490 | mod_timer(&priv->poll_timer, jiffies + 1); | |
491 | ||
943c246e | 492 | netif_tx_unlock(dev); |
57ce41d1 EC |
493 | } |
494 | ||
495 | void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr) | |
496 | { | |
943c246e RD |
497 | struct ipoib_dev_priv *priv = netdev_priv(dev_ptr); |
498 | ||
499 | mod_timer(&priv->poll_timer, jiffies); | |
57ce41d1 EC |
500 | } |
501 | ||
1da177e4 LT |
502 | static inline int post_send(struct ipoib_dev_priv *priv, |
503 | unsigned int wr_id, | |
504 | struct ib_ah *address, u32 qpn, | |
40ca1988 EC |
505 | struct ipoib_tx_buf *tx_req, |
506 | void *head, int hlen) | |
1da177e4 LT |
507 | { |
508 | struct ib_send_wr *bad_wr; | |
40ca1988 EC |
509 | int i, off; |
510 | struct sk_buff *skb = tx_req->skb; | |
511 | skb_frag_t *frags = skb_shinfo(skb)->frags; | |
512 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
513 | u64 *mapping = tx_req->mapping; | |
514 | ||
515 | if (skb_headlen(skb)) { | |
516 | priv->tx_sge[0].addr = mapping[0]; | |
517 | priv->tx_sge[0].length = skb_headlen(skb); | |
518 | off = 1; | |
519 | } else | |
520 | off = 0; | |
1da177e4 | 521 | |
7143740d | 522 | for (i = 0; i < nr_frags; ++i) { |
40ca1988 | 523 | priv->tx_sge[i + off].addr = mapping[i + off]; |
9e903e08 | 524 | priv->tx_sge[i + off].length = skb_frag_size(&frags[i]); |
7143740d | 525 | } |
40ca1988 | 526 | priv->tx_wr.num_sge = nr_frags + off; |
7143740d EC |
527 | priv->tx_wr.wr_id = wr_id; |
528 | priv->tx_wr.wr.ud.remote_qpn = qpn; | |
529 | priv->tx_wr.wr.ud.ah = address; | |
1da177e4 | 530 | |
40ca1988 EC |
531 | if (head) { |
532 | priv->tx_wr.wr.ud.mss = skb_shinfo(skb)->gso_size; | |
533 | priv->tx_wr.wr.ud.header = head; | |
534 | priv->tx_wr.wr.ud.hlen = hlen; | |
535 | priv->tx_wr.opcode = IB_WR_LSO; | |
536 | } else | |
537 | priv->tx_wr.opcode = IB_WR_SEND; | |
538 | ||
1da177e4 LT |
539 | return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr); |
540 | } | |
541 | ||
542 | void ipoib_send(struct net_device *dev, struct sk_buff *skb, | |
543 | struct ipoib_ah *address, u32 qpn) | |
544 | { | |
545 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1993d683 | 546 | struct ipoib_tx_buf *tx_req; |
a48f509b | 547 | int hlen, rc; |
40ca1988 EC |
548 | void *phead; |
549 | ||
550 | if (skb_is_gso(skb)) { | |
551 | hlen = skb_transport_offset(skb) + tcp_hdrlen(skb); | |
552 | phead = skb->data; | |
553 | if (unlikely(!skb_pull(skb, hlen))) { | |
554 | ipoib_warn(priv, "linear data too small\n"); | |
555 | ++dev->stats.tx_dropped; | |
556 | ++dev->stats.tx_errors; | |
557 | dev_kfree_skb_any(skb); | |
558 | return; | |
559 | } | |
560 | } else { | |
561 | if (unlikely(skb->len > priv->mcast_mtu + IPOIB_ENCAP_LEN)) { | |
562 | ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n", | |
563 | skb->len, priv->mcast_mtu + IPOIB_ENCAP_LEN); | |
564 | ++dev->stats.tx_dropped; | |
565 | ++dev->stats.tx_errors; | |
566 | ipoib_cm_skb_too_long(dev, skb, priv->mcast_mtu); | |
567 | return; | |
568 | } | |
569 | phead = NULL; | |
570 | hlen = 0; | |
1da177e4 LT |
571 | } |
572 | ||
573 | ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n", | |
574 | skb->len, address, qpn); | |
575 | ||
576 | /* | |
577 | * We put the skb into the tx_ring _before_ we call post_send() | |
578 | * because it's entirely possible that the completion handler will | |
579 | * run before we execute anything after the post_send(). That | |
580 | * means we have to make sure everything is properly recorded and | |
581 | * our state is consistent before we call post_send(). | |
582 | */ | |
0f485251 | 583 | tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)]; |
1da177e4 | 584 | tx_req->skb = skb; |
7143740d | 585 | if (unlikely(ipoib_dma_map_tx(priv->ca, tx_req))) { |
de903512 | 586 | ++dev->stats.tx_errors; |
73fbe8be RD |
587 | dev_kfree_skb_any(skb); |
588 | return; | |
589 | } | |
1da177e4 | 590 | |
6046136c EC |
591 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
592 | priv->tx_wr.send_flags |= IB_SEND_IP_CSUM; | |
593 | else | |
594 | priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM; | |
595 | ||
57ce41d1 EC |
596 | if (++priv->tx_outstanding == ipoib_sendq_size) { |
597 | ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n"); | |
598 | if (ib_req_notify_cq(priv->send_cq, IB_CQ_NEXT_COMP)) | |
599 | ipoib_warn(priv, "request notify on send CQ failed\n"); | |
600 | netif_stop_queue(dev); | |
601 | } | |
602 | ||
7e5a90c2 SP |
603 | skb_orphan(skb); |
604 | skb_dst_drop(skb); | |
605 | ||
a48f509b OG |
606 | rc = post_send(priv, priv->tx_head & (ipoib_sendq_size - 1), |
607 | address->ah, qpn, tx_req, phead, hlen); | |
608 | if (unlikely(rc)) { | |
609 | ipoib_warn(priv, "post_send failed, error %d\n", rc); | |
de903512 | 610 | ++dev->stats.tx_errors; |
57ce41d1 | 611 | --priv->tx_outstanding; |
7143740d | 612 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
1da177e4 | 613 | dev_kfree_skb_any(skb); |
57ce41d1 EC |
614 | if (netif_queue_stopped(dev)) |
615 | netif_wake_queue(dev); | |
1da177e4 LT |
616 | } else { |
617 | dev->trans_start = jiffies; | |
618 | ||
619 | address->last_send = priv->tx_head; | |
620 | ++priv->tx_head; | |
1da177e4 | 621 | } |
f56bcd80 EC |
622 | |
623 | if (unlikely(priv->tx_outstanding > MAX_SEND_CQE)) | |
57ce41d1 EC |
624 | while (poll_tx(priv)) |
625 | ; /* nothing */ | |
1da177e4 LT |
626 | } |
627 | ||
628 | static void __ipoib_reap_ah(struct net_device *dev) | |
629 | { | |
630 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
631 | struct ipoib_ah *ah, *tah; | |
632 | LIST_HEAD(remove_list); | |
943c246e RD |
633 | unsigned long flags; |
634 | ||
635 | netif_tx_lock_bh(dev); | |
636 | spin_lock_irqsave(&priv->lock, flags); | |
1da177e4 | 637 | |
1da177e4 | 638 | list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list) |
2181858b | 639 | if ((int) priv->tx_tail - (int) ah->last_send >= 0) { |
1da177e4 | 640 | list_del(&ah->list); |
31c02e21 RD |
641 | ib_destroy_ah(ah->ah); |
642 | kfree(ah); | |
1da177e4 | 643 | } |
943c246e RD |
644 | |
645 | spin_unlock_irqrestore(&priv->lock, flags); | |
646 | netif_tx_unlock_bh(dev); | |
1da177e4 LT |
647 | } |
648 | ||
c4028958 | 649 | void ipoib_reap_ah(struct work_struct *work) |
1da177e4 | 650 | { |
c4028958 DH |
651 | struct ipoib_dev_priv *priv = |
652 | container_of(work, struct ipoib_dev_priv, ah_reap_task.work); | |
653 | struct net_device *dev = priv->dev; | |
1da177e4 LT |
654 | |
655 | __ipoib_reap_ah(dev); | |
656 | ||
657 | if (!test_bit(IPOIB_STOP_REAPER, &priv->flags)) | |
0306eda2 | 658 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, |
69fc507a | 659 | round_jiffies_relative(HZ)); |
1da177e4 LT |
660 | } |
661 | ||
57ce41d1 EC |
662 | static void ipoib_ib_tx_timer_func(unsigned long ctx) |
663 | { | |
664 | drain_tx_cq((struct net_device *)ctx); | |
665 | } | |
666 | ||
a84544a4 | 667 | int ipoib_ib_dev_open(struct net_device *dev, int flush) |
1da177e4 LT |
668 | { |
669 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
670 | int ret; | |
671 | ||
dd57c930 AE |
672 | ipoib_pkey_dev_check_presence(dev); |
673 | ||
674 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
675 | ipoib_warn(priv, "P_Key 0x%04x is %s\n", priv->pkey, | |
676 | (!(priv->pkey & 0x7fff) ? "Invalid" : "not found")); | |
26bbf13c YE |
677 | return -1; |
678 | } | |
26bbf13c | 679 | |
5b6810e0 | 680 | ret = ipoib_init_qp(dev); |
1da177e4 | 681 | if (ret) { |
5b6810e0 | 682 | ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret); |
1da177e4 LT |
683 | return -1; |
684 | } | |
685 | ||
686 | ret = ipoib_ib_post_receives(dev); | |
687 | if (ret) { | |
688 | ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret); | |
c2bb5628 | 689 | goto dev_stop; |
1da177e4 LT |
690 | } |
691 | ||
839fcaba MT |
692 | ret = ipoib_cm_dev_open(dev); |
693 | if (ret) { | |
24bd1e4e | 694 | ipoib_warn(priv, "ipoib_cm_dev_open returned %d\n", ret); |
c2bb5628 | 695 | goto dev_stop; |
839fcaba MT |
696 | } |
697 | ||
1da177e4 | 698 | clear_bit(IPOIB_STOP_REAPER, &priv->flags); |
0306eda2 | 699 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, |
69fc507a | 700 | round_jiffies_relative(HZ)); |
1da177e4 | 701 | |
e028cc55 YE |
702 | if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
703 | napi_enable(&priv->napi); | |
7a343d4c | 704 | |
1da177e4 | 705 | return 0; |
c2bb5628 ES |
706 | dev_stop: |
707 | if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) | |
708 | napi_enable(&priv->napi); | |
a84544a4 | 709 | ipoib_ib_dev_stop(dev, flush); |
c2bb5628 | 710 | return -1; |
1da177e4 LT |
711 | } |
712 | ||
db84f880 | 713 | void ipoib_pkey_dev_check_presence(struct net_device *dev) |
7a343d4c LA |
714 | { |
715 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
7a343d4c | 716 | |
dd57c930 AE |
717 | if (!(priv->pkey & 0x7fff) || |
718 | ib_find_pkey(priv->ca, priv->port, priv->pkey, | |
719 | &priv->pkey_index)) | |
7a343d4c LA |
720 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); |
721 | else | |
722 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
723 | } | |
724 | ||
1da177e4 LT |
725 | int ipoib_ib_dev_up(struct net_device *dev) |
726 | { | |
727 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
728 | ||
7a343d4c LA |
729 | ipoib_pkey_dev_check_presence(dev); |
730 | ||
731 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
732 | ipoib_dbg(priv, "PKEY is not assigned.\n"); | |
733 | return 0; | |
734 | } | |
735 | ||
1da177e4 LT |
736 | set_bit(IPOIB_FLAG_OPER_UP, &priv->flags); |
737 | ||
738 | return ipoib_mcast_start_thread(dev); | |
739 | } | |
740 | ||
a84544a4 | 741 | int ipoib_ib_dev_down(struct net_device *dev, int flush) |
1da177e4 LT |
742 | { |
743 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
744 | ||
745 | ipoib_dbg(priv, "downing ib_dev\n"); | |
746 | ||
747 | clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags); | |
748 | netif_carrier_off(dev); | |
749 | ||
4e0ab200 | 750 | ipoib_mcast_stop_thread(dev, flush); |
1da177e4 LT |
751 | ipoib_mcast_dev_flush(dev); |
752 | ||
1da177e4 LT |
753 | ipoib_flush_paths(dev); |
754 | ||
755 | return 0; | |
756 | } | |
757 | ||
758 | static int recvs_pending(struct net_device *dev) | |
759 | { | |
760 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
761 | int pending = 0; | |
762 | int i; | |
763 | ||
0f485251 | 764 | for (i = 0; i < ipoib_recvq_size; ++i) |
1da177e4 LT |
765 | if (priv->rx_ring[i].skb) |
766 | ++pending; | |
767 | ||
768 | return pending; | |
769 | } | |
770 | ||
2dfbfc37 MT |
771 | void ipoib_drain_cq(struct net_device *dev) |
772 | { | |
773 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
774 | int i, n; | |
943c246e RD |
775 | |
776 | /* | |
777 | * We call completion handling routines that expect to be | |
778 | * called from the BH-disabled NAPI poll context, so disable | |
779 | * BHs here too. | |
780 | */ | |
781 | local_bh_disable(); | |
782 | ||
2dfbfc37 | 783 | do { |
f56bcd80 | 784 | n = ib_poll_cq(priv->recv_cq, IPOIB_NUM_WC, priv->ibwc); |
2dfbfc37 | 785 | for (i = 0; i < n; ++i) { |
ce423ef5 RD |
786 | /* |
787 | * Convert any successful completions to flush | |
788 | * errors to avoid passing packets up the | |
789 | * stack after bringing the device down. | |
790 | */ | |
791 | if (priv->ibwc[i].status == IB_WC_SUCCESS) | |
792 | priv->ibwc[i].status = IB_WC_WR_FLUSH_ERR; | |
793 | ||
1b524963 MT |
794 | if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) { |
795 | if (priv->ibwc[i].wr_id & IPOIB_OP_CM) | |
796 | ipoib_cm_handle_rx_wc(dev, priv->ibwc + i); | |
797 | else | |
798 | ipoib_ib_handle_rx_wc(dev, priv->ibwc + i); | |
f56bcd80 EC |
799 | } else |
800 | ipoib_cm_handle_tx_wc(dev, priv->ibwc + i); | |
2dfbfc37 MT |
801 | } |
802 | } while (n == IPOIB_NUM_WC); | |
f56bcd80 EC |
803 | |
804 | while (poll_tx(priv)) | |
805 | ; /* nothing */ | |
943c246e RD |
806 | |
807 | local_bh_enable(); | |
2dfbfc37 MT |
808 | } |
809 | ||
a84544a4 | 810 | int ipoib_ib_dev_stop(struct net_device *dev, int flush) |
1da177e4 LT |
811 | { |
812 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
813 | struct ib_qp_attr qp_attr; | |
1da177e4 | 814 | unsigned long begin; |
1993d683 | 815 | struct ipoib_tx_buf *tx_req; |
2dfbfc37 | 816 | int i; |
1da177e4 | 817 | |
e028cc55 YE |
818 | if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
819 | napi_disable(&priv->napi); | |
7a343d4c | 820 | |
839fcaba MT |
821 | ipoib_cm_dev_stop(dev); |
822 | ||
3bc12e75 RD |
823 | /* |
824 | * Move our QP to the error state and then reinitialize in | |
825 | * when all work requests have completed or have been flushed. | |
826 | */ | |
1da177e4 | 827 | qp_attr.qp_state = IB_QPS_ERR; |
3bc12e75 | 828 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
829 | ipoib_warn(priv, "Failed to modify QP to ERROR state\n"); |
830 | ||
831 | /* Wait for all sends and receives to complete */ | |
832 | begin = jiffies; | |
833 | ||
834 | while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) { | |
835 | if (time_after(jiffies, begin + 5 * HZ)) { | |
836 | ipoib_warn(priv, "timing out; %d sends %d receives not completed\n", | |
837 | priv->tx_head - priv->tx_tail, recvs_pending(dev)); | |
838 | ||
839 | /* | |
840 | * assume the HW is wedged and just free up | |
841 | * all our pending work requests. | |
842 | */ | |
2181858b | 843 | while ((int) priv->tx_tail - (int) priv->tx_head < 0) { |
1da177e4 | 844 | tx_req = &priv->tx_ring[priv->tx_tail & |
0f485251 | 845 | (ipoib_sendq_size - 1)]; |
7143740d | 846 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
1da177e4 LT |
847 | dev_kfree_skb_any(tx_req->skb); |
848 | ++priv->tx_tail; | |
1b524963 | 849 | --priv->tx_outstanding; |
1da177e4 LT |
850 | } |
851 | ||
37ccf9df RC |
852 | for (i = 0; i < ipoib_recvq_size; ++i) { |
853 | struct ipoib_rx_buf *rx_req; | |
854 | ||
855 | rx_req = &priv->rx_ring[i]; | |
856 | if (!rx_req->skb) | |
857 | continue; | |
bc7b3a36 SM |
858 | ipoib_ud_dma_unmap_rx(priv, |
859 | priv->rx_ring[i].mapping); | |
37ccf9df RC |
860 | dev_kfree_skb_any(rx_req->skb); |
861 | rx_req->skb = NULL; | |
862 | } | |
1da177e4 LT |
863 | |
864 | goto timeout; | |
865 | } | |
866 | ||
2dfbfc37 | 867 | ipoib_drain_cq(dev); |
8d1cc86a | 868 | |
1da177e4 LT |
869 | msleep(1); |
870 | } | |
871 | ||
872 | ipoib_dbg(priv, "All sends and receives done.\n"); | |
873 | ||
874 | timeout: | |
57ce41d1 | 875 | del_timer_sync(&priv->poll_timer); |
1da177e4 | 876 | qp_attr.qp_state = IB_QPS_RESET; |
3bc12e75 | 877 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
878 | ipoib_warn(priv, "Failed to modify QP to RESET state\n"); |
879 | ||
880 | /* Wait for all AHs to be reaped */ | |
881 | set_bit(IPOIB_STOP_REAPER, &priv->flags); | |
882 | cancel_delayed_work(&priv->ah_reap_task); | |
a84544a4 | 883 | if (flush) |
0306eda2 | 884 | flush_workqueue(ipoib_workqueue); |
1da177e4 LT |
885 | |
886 | begin = jiffies; | |
887 | ||
888 | while (!list_empty(&priv->dead_ahs)) { | |
889 | __ipoib_reap_ah(dev); | |
890 | ||
891 | if (time_after(jiffies, begin + HZ)) { | |
892 | ipoib_warn(priv, "timing out; will leak address handles\n"); | |
893 | break; | |
894 | } | |
895 | ||
896 | msleep(1); | |
897 | } | |
898 | ||
f56bcd80 | 899 | ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP); |
8d1cc86a | 900 | |
1da177e4 LT |
901 | return 0; |
902 | } | |
903 | ||
904 | int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port) | |
905 | { | |
906 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
907 | ||
908 | priv->ca = ca; | |
909 | priv->port = port; | |
910 | priv->qp = NULL; | |
911 | ||
912 | if (ipoib_transport_dev_init(dev, ca)) { | |
913 | printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name); | |
914 | return -ENODEV; | |
915 | } | |
916 | ||
2767840a RD |
917 | setup_timer(&priv->poll_timer, ipoib_ib_tx_timer_func, |
918 | (unsigned long) dev); | |
919 | ||
1da177e4 | 920 | if (dev->flags & IFF_UP) { |
a84544a4 | 921 | if (ipoib_ib_dev_open(dev, 1)) { |
1da177e4 LT |
922 | ipoib_transport_dev_cleanup(dev); |
923 | return -ENODEV; | |
924 | } | |
925 | } | |
926 | ||
927 | return 0; | |
928 | } | |
929 | ||
c2904141 ES |
930 | /* |
931 | * Takes whatever value which is in pkey index 0 and updates priv->pkey | |
932 | * returns 0 if the pkey value was changed. | |
933 | */ | |
934 | static inline int update_parent_pkey(struct ipoib_dev_priv *priv) | |
935 | { | |
936 | int result; | |
937 | u16 prev_pkey; | |
938 | ||
939 | prev_pkey = priv->pkey; | |
940 | result = ib_query_pkey(priv->ca, priv->port, 0, &priv->pkey); | |
941 | if (result) { | |
942 | ipoib_warn(priv, "ib_query_pkey port %d failed (ret = %d)\n", | |
943 | priv->port, result); | |
944 | return result; | |
945 | } | |
946 | ||
947 | priv->pkey |= 0x8000; | |
948 | ||
949 | if (prev_pkey != priv->pkey) { | |
950 | ipoib_dbg(priv, "pkey changed from 0x%x to 0x%x\n", | |
951 | prev_pkey, priv->pkey); | |
952 | /* | |
953 | * Update the pkey in the broadcast address, while making sure to set | |
954 | * the full membership bit, so that we join the right broadcast group. | |
955 | */ | |
956 | priv->dev->broadcast[8] = priv->pkey >> 8; | |
957 | priv->dev->broadcast[9] = priv->pkey & 0xff; | |
958 | return 0; | |
959 | } | |
960 | ||
961 | return 1; | |
962 | } | |
dd57c930 AE |
963 | /* |
964 | * returns 0 if pkey value was found in a different slot. | |
965 | */ | |
966 | static inline int update_child_pkey(struct ipoib_dev_priv *priv) | |
967 | { | |
968 | u16 old_index = priv->pkey_index; | |
969 | ||
970 | priv->pkey_index = 0; | |
971 | ipoib_pkey_dev_check_presence(priv->dev); | |
972 | ||
973 | if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags) && | |
974 | (old_index == priv->pkey_index)) | |
975 | return 1; | |
976 | return 0; | |
977 | } | |
c2904141 | 978 | |
ee1e2c82 MS |
979 | static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, |
980 | enum ipoib_flush_level level) | |
1da177e4 | 981 | { |
26bbf13c | 982 | struct ipoib_dev_priv *cpriv; |
c4028958 | 983 | struct net_device *dev = priv->dev; |
c2904141 | 984 | int result; |
26bbf13c | 985 | |
f47944cc | 986 | down_read(&priv->vlan_rwsem); |
1da177e4 | 987 | |
26bbf13c YE |
988 | /* |
989 | * Flush any child interfaces too -- they might be up even if | |
990 | * the parent is down. | |
991 | */ | |
992 | list_for_each_entry(cpriv, &priv->child_intfs, list) | |
ee1e2c82 | 993 | __ipoib_ib_dev_flush(cpriv, level); |
26bbf13c | 994 | |
f47944cc | 995 | up_read(&priv->vlan_rwsem); |
26bbf13c | 996 | |
dd57c930 AE |
997 | if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags) && |
998 | level != IPOIB_FLUSH_HEAVY) { | |
7a343d4c LA |
999 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n"); |
1000 | return; | |
1001 | } | |
1002 | ||
1003 | if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { | |
dd57c930 AE |
1004 | /* interface is down. update pkey and leave. */ |
1005 | if (level == IPOIB_FLUSH_HEAVY) { | |
1006 | if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) | |
1007 | update_parent_pkey(priv); | |
1008 | else | |
1009 | update_child_pkey(priv); | |
1010 | } | |
7a343d4c | 1011 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n"); |
1da177e4 | 1012 | return; |
7a343d4c | 1013 | } |
1da177e4 | 1014 | |
ee1e2c82 | 1015 | if (level == IPOIB_FLUSH_HEAVY) { |
c2904141 ES |
1016 | /* child devices chase their origin pkey value, while non-child |
1017 | * (parent) devices should always takes what present in pkey index 0 | |
1018 | */ | |
1019 | if (test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { | |
dd57c930 AE |
1020 | result = update_child_pkey(priv); |
1021 | if (result) { | |
1022 | /* restart QP only if P_Key index is changed */ | |
c2904141 | 1023 | ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n"); |
9fdd5e5b | 1024 | return; |
c2904141 | 1025 | } |
dd57c930 | 1026 | |
c2904141 ES |
1027 | } else { |
1028 | result = update_parent_pkey(priv); | |
1029 | /* restart QP only if P_Key value changed */ | |
1030 | if (result) { | |
1031 | ipoib_dbg(priv, "Not flushing - P_Key value not changed.\n"); | |
1032 | return; | |
1033 | } | |
26bbf13c | 1034 | } |
26bbf13c YE |
1035 | } |
1036 | ||
ee1e2c82 MS |
1037 | if (level == IPOIB_FLUSH_LIGHT) { |
1038 | ipoib_mark_paths_invalid(dev); | |
1039 | ipoib_mcast_dev_flush(dev); | |
1040 | } | |
1da177e4 | 1041 | |
ee1e2c82 | 1042 | if (level >= IPOIB_FLUSH_NORMAL) |
a84544a4 | 1043 | ipoib_ib_dev_down(dev, 0); |
1da177e4 | 1044 | |
ee1e2c82 | 1045 | if (level == IPOIB_FLUSH_HEAVY) { |
dd57c930 | 1046 | if (test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
a84544a4 RD |
1047 | ipoib_ib_dev_stop(dev, 0); |
1048 | if (ipoib_ib_dev_open(dev, 0) != 0) | |
dd57c930 AE |
1049 | return; |
1050 | if (netif_queue_stopped(dev)) | |
1051 | netif_start_queue(dev); | |
26bbf13c YE |
1052 | } |
1053 | ||
1da177e4 LT |
1054 | /* |
1055 | * The device could have been brought down between the start and when | |
1056 | * we get here, don't bring it back up if it's not configured up | |
1057 | */ | |
5ccd0255 | 1058 | if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { |
ee1e2c82 MS |
1059 | if (level >= IPOIB_FLUSH_NORMAL) |
1060 | ipoib_ib_dev_up(dev); | |
c4028958 | 1061 | ipoib_mcast_restart_task(&priv->restart_task); |
5ccd0255 | 1062 | } |
26bbf13c | 1063 | } |
1da177e4 | 1064 | |
ee1e2c82 MS |
1065 | void ipoib_ib_dev_flush_light(struct work_struct *work) |
1066 | { | |
1067 | struct ipoib_dev_priv *priv = | |
1068 | container_of(work, struct ipoib_dev_priv, flush_light); | |
1069 | ||
1070 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_LIGHT); | |
1071 | } | |
1072 | ||
1073 | void ipoib_ib_dev_flush_normal(struct work_struct *work) | |
26bbf13c YE |
1074 | { |
1075 | struct ipoib_dev_priv *priv = | |
ee1e2c82 | 1076 | container_of(work, struct ipoib_dev_priv, flush_normal); |
4f71055a | 1077 | |
ee1e2c82 | 1078 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_NORMAL); |
26bbf13c | 1079 | } |
4f71055a | 1080 | |
ee1e2c82 | 1081 | void ipoib_ib_dev_flush_heavy(struct work_struct *work) |
26bbf13c YE |
1082 | { |
1083 | struct ipoib_dev_priv *priv = | |
ee1e2c82 | 1084 | container_of(work, struct ipoib_dev_priv, flush_heavy); |
26bbf13c | 1085 | |
ee1e2c82 | 1086 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_HEAVY); |
1da177e4 LT |
1087 | } |
1088 | ||
1089 | void ipoib_ib_dev_cleanup(struct net_device *dev) | |
1090 | { | |
1091 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1092 | ||
1093 | ipoib_dbg(priv, "cleaning up ib_dev\n"); | |
a39c52ab ES |
1094 | /* |
1095 | * We must make sure there are no more (path) completions | |
1096 | * that may wish to touch priv fields that are no longer valid | |
1097 | */ | |
1098 | ipoib_flush_paths(dev); | |
1da177e4 | 1099 | |
4e0ab200 | 1100 | ipoib_mcast_stop_thread(dev, 1); |
988bd503 | 1101 | ipoib_mcast_dev_flush(dev); |
1da177e4 LT |
1102 | |
1103 | ipoib_transport_dev_cleanup(dev); | |
1104 | } | |
1105 | ||
1da177e4 | 1106 |