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 | { | |
a44878d1 ES |
97 | ib_dma_unmap_single(priv->ca, mapping[0], |
98 | IPOIB_UD_BUF_SIZE(priv->max_ib_mtu), | |
99 | DMA_FROM_DEVICE); | |
bc7b3a36 SM |
100 | } |
101 | ||
1993d683 | 102 | static int ipoib_ib_post_receive(struct net_device *dev, int id) |
1da177e4 | 103 | { |
1993d683 | 104 | struct ipoib_dev_priv *priv = netdev_priv(dev); |
1da177e4 | 105 | struct ib_recv_wr *bad_wr; |
1993d683 RD |
106 | int ret; |
107 | ||
bc7b3a36 SM |
108 | priv->rx_wr.wr_id = id | IPOIB_OP_RECV; |
109 | priv->rx_sge[0].addr = priv->rx_ring[id].mapping[0]; | |
110 | priv->rx_sge[1].addr = priv->rx_ring[id].mapping[1]; | |
1993d683 | 111 | |
1993d683 | 112 | |
bc7b3a36 | 113 | ret = ib_post_recv(priv->qp, &priv->rx_wr, &bad_wr); |
1993d683 RD |
114 | if (unlikely(ret)) { |
115 | ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret); | |
bc7b3a36 | 116 | ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[id].mapping); |
1993d683 RD |
117 | dev_kfree_skb_any(priv->rx_ring[id].skb); |
118 | priv->rx_ring[id].skb = NULL; | |
119 | } | |
1da177e4 | 120 | |
1993d683 | 121 | return ret; |
1da177e4 LT |
122 | } |
123 | ||
bc7b3a36 | 124 | static struct sk_buff *ipoib_alloc_rx_skb(struct net_device *dev, int id) |
1da177e4 LT |
125 | { |
126 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
127 | struct sk_buff *skb; | |
bc7b3a36 SM |
128 | int buf_size; |
129 | u64 *mapping; | |
1da177e4 | 130 | |
a44878d1 | 131 | buf_size = IPOIB_UD_BUF_SIZE(priv->max_ib_mtu); |
bc7b3a36 | 132 | |
a44878d1 | 133 | skb = dev_alloc_skb(buf_size + IPOIB_ENCAP_LEN); |
bc7b3a36 SM |
134 | if (unlikely(!skb)) |
135 | return NULL; | |
1993d683 RD |
136 | |
137 | /* | |
138 | * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte | |
139 | * header. So we need 4 more bytes to get to 48 and align the | |
140 | * IP header to a multiple of 16. | |
141 | */ | |
142 | skb_reserve(skb, 4); | |
143 | ||
bc7b3a36 SM |
144 | mapping = priv->rx_ring[id].mapping; |
145 | mapping[0] = ib_dma_map_single(priv->ca, skb->data, buf_size, | |
146 | DMA_FROM_DEVICE); | |
147 | if (unlikely(ib_dma_mapping_error(priv->ca, mapping[0]))) | |
148 | goto error; | |
149 | ||
bc7b3a36 SM |
150 | priv->rx_ring[id].skb = skb; |
151 | return skb; | |
bc7b3a36 SM |
152 | error: |
153 | dev_kfree_skb_any(skb); | |
154 | return NULL; | |
1da177e4 LT |
155 | } |
156 | ||
157 | static int ipoib_ib_post_receives(struct net_device *dev) | |
158 | { | |
159 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
160 | int i; | |
161 | ||
0f485251 | 162 | for (i = 0; i < ipoib_recvq_size; ++i) { |
bc7b3a36 | 163 | if (!ipoib_alloc_rx_skb(dev, i)) { |
1993d683 RD |
164 | ipoib_warn(priv, "failed to allocate receive buffer %d\n", i); |
165 | return -ENOMEM; | |
166 | } | |
1da177e4 LT |
167 | if (ipoib_ib_post_receive(dev, i)) { |
168 | ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i); | |
169 | return -EIO; | |
170 | } | |
171 | } | |
172 | ||
173 | return 0; | |
174 | } | |
175 | ||
2439a6e6 | 176 | static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) |
1da177e4 LT |
177 | { |
178 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
2439a6e6 RD |
179 | unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV; |
180 | struct sk_buff *skb; | |
bc7b3a36 | 181 | u64 mapping[IPOIB_UD_RX_SG]; |
fed1db33 | 182 | union ib_gid *dgid; |
1da177e4 | 183 | |
a89875fc RD |
184 | ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n", |
185 | wr_id, wc->status); | |
1da177e4 | 186 | |
2439a6e6 RD |
187 | if (unlikely(wr_id >= ipoib_recvq_size)) { |
188 | ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n", | |
189 | wr_id, ipoib_recvq_size); | |
190 | return; | |
191 | } | |
192 | ||
193 | skb = priv->rx_ring[wr_id].skb; | |
2439a6e6 RD |
194 | |
195 | if (unlikely(wc->status != IB_WC_SUCCESS)) { | |
196 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
197 | ipoib_warn(priv, "failed recv event " | |
198 | "(status=%d, wrid=%d vend_err %x)\n", | |
199 | wc->status, wr_id, wc->vendor_err); | |
bc7b3a36 | 200 | ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[wr_id].mapping); |
2439a6e6 RD |
201 | dev_kfree_skb_any(skb); |
202 | priv->rx_ring[wr_id].skb = NULL; | |
203 | return; | |
204 | } | |
1da177e4 | 205 | |
1b844afe RD |
206 | /* |
207 | * Drop packets that this interface sent, ie multicast packets | |
208 | * that the HCA has replicated. | |
209 | */ | |
210 | if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num) | |
211 | goto repost; | |
212 | ||
bc7b3a36 SM |
213 | memcpy(mapping, priv->rx_ring[wr_id].mapping, |
214 | IPOIB_UD_RX_SG * sizeof *mapping); | |
215 | ||
2439a6e6 RD |
216 | /* |
217 | * If we can't allocate a new RX buffer, dump | |
218 | * this packet and reuse the old buffer. | |
219 | */ | |
bc7b3a36 | 220 | if (unlikely(!ipoib_alloc_rx_skb(dev, wr_id))) { |
de903512 | 221 | ++dev->stats.rx_dropped; |
2439a6e6 RD |
222 | goto repost; |
223 | } | |
224 | ||
225 | ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n", | |
226 | wc->byte_len, wc->slid); | |
227 | ||
bc7b3a36 | 228 | ipoib_ud_dma_unmap_rx(priv, mapping); |
a44878d1 ES |
229 | |
230 | skb_put(skb, wc->byte_len); | |
2439a6e6 | 231 | |
fed1db33 CL |
232 | /* First byte of dgid signals multicast when 0xff */ |
233 | dgid = &((struct ib_grh *)skb->data)->dgid; | |
234 | ||
235 | if (!(wc->wc_flags & IB_WC_GRH) || dgid->raw[0] != 0xff) | |
236 | skb->pkt_type = PACKET_HOST; | |
237 | else if (memcmp(dgid, dev->broadcast + 4, sizeof(union ib_gid)) == 0) | |
238 | skb->pkt_type = PACKET_BROADCAST; | |
239 | else | |
240 | skb->pkt_type = PACKET_MULTICAST; | |
241 | ||
2439a6e6 RD |
242 | skb_pull(skb, IB_GRH_BYTES); |
243 | ||
1b844afe RD |
244 | skb->protocol = ((struct ipoib_header *) skb->data)->proto; |
245 | skb_reset_mac_header(skb); | |
246 | skb_pull(skb, IPOIB_ENCAP_LEN); | |
247 | ||
de903512 RD |
248 | ++dev->stats.rx_packets; |
249 | dev->stats.rx_bytes += skb->len; | |
1b844afe RD |
250 | |
251 | skb->dev = dev; | |
d927d505 OG |
252 | if ((dev->features & NETIF_F_RXCSUM) && |
253 | likely(wc->wc_flags & IB_WC_IP_CSUM_OK)) | |
6046136c EC |
254 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
255 | ||
8ae31e5b | 256 | napi_gro_receive(&priv->napi, skb); |
1da177e4 | 257 | |
2439a6e6 RD |
258 | repost: |
259 | if (unlikely(ipoib_ib_post_receive(dev, wr_id))) | |
260 | ipoib_warn(priv, "ipoib_ib_post_receive failed " | |
261 | "for buf %d\n", wr_id); | |
262 | } | |
1da177e4 | 263 | |
c4268778 | 264 | int ipoib_dma_map_tx(struct ib_device *ca, struct ipoib_tx_buf *tx_req) |
7143740d EC |
265 | { |
266 | struct sk_buff *skb = tx_req->skb; | |
267 | u64 *mapping = tx_req->mapping; | |
268 | int i; | |
40ca1988 | 269 | int off; |
7143740d | 270 | |
40ca1988 EC |
271 | if (skb_headlen(skb)) { |
272 | mapping[0] = ib_dma_map_single(ca, skb->data, skb_headlen(skb), | |
273 | DMA_TO_DEVICE); | |
274 | if (unlikely(ib_dma_mapping_error(ca, mapping[0]))) | |
275 | return -EIO; | |
276 | ||
277 | off = 1; | |
278 | } else | |
279 | off = 0; | |
7143740d EC |
280 | |
281 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
9e903e08 | 282 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
5581be3b IC |
283 | mapping[i + off] = ib_dma_map_page(ca, |
284 | skb_frag_page(frag), | |
9e903e08 | 285 | frag->page_offset, skb_frag_size(frag), |
7143740d | 286 | DMA_TO_DEVICE); |
40ca1988 | 287 | if (unlikely(ib_dma_mapping_error(ca, mapping[i + off]))) |
7143740d EC |
288 | goto partial_error; |
289 | } | |
290 | return 0; | |
291 | ||
292 | partial_error: | |
7143740d | 293 | for (; i > 0; --i) { |
9e903e08 ED |
294 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; |
295 | ||
296 | ib_dma_unmap_page(ca, mapping[i - !off], skb_frag_size(frag), DMA_TO_DEVICE); | |
7143740d | 297 | } |
40ca1988 EC |
298 | |
299 | if (off) | |
300 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
301 | ||
7143740d EC |
302 | return -EIO; |
303 | } | |
304 | ||
c4268778 YS |
305 | void ipoib_dma_unmap_tx(struct ipoib_dev_priv *priv, |
306 | struct ipoib_tx_buf *tx_req) | |
7143740d EC |
307 | { |
308 | struct sk_buff *skb = tx_req->skb; | |
309 | u64 *mapping = tx_req->mapping; | |
310 | int i; | |
40ca1988 | 311 | int off; |
7143740d | 312 | |
40ca1988 | 313 | if (skb_headlen(skb)) { |
c4268778 YS |
314 | ib_dma_unmap_single(priv->ca, mapping[0], skb_headlen(skb), |
315 | DMA_TO_DEVICE); | |
40ca1988 EC |
316 | off = 1; |
317 | } else | |
318 | off = 0; | |
7143740d EC |
319 | |
320 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
9e903e08 ED |
321 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
322 | ||
c4268778 YS |
323 | ib_dma_unmap_page(priv->ca, mapping[i + off], |
324 | skb_frag_size(frag), DMA_TO_DEVICE); | |
7143740d EC |
325 | } |
326 | } | |
327 | ||
2c010730 ES |
328 | /* |
329 | * As the result of a completion error the QP Can be transferred to SQE states. | |
330 | * The function checks if the (send)QP is in SQE state and | |
331 | * moves it back to RTS state, that in order to have it functional again. | |
332 | */ | |
333 | static void ipoib_qp_state_validate_work(struct work_struct *work) | |
334 | { | |
335 | struct ipoib_qp_state_validate *qp_work = | |
336 | container_of(work, struct ipoib_qp_state_validate, work); | |
337 | ||
338 | struct ipoib_dev_priv *priv = qp_work->priv; | |
339 | struct ib_qp_attr qp_attr; | |
340 | struct ib_qp_init_attr query_init_attr; | |
341 | int ret; | |
342 | ||
343 | ret = ib_query_qp(priv->qp, &qp_attr, IB_QP_STATE, &query_init_attr); | |
344 | if (ret) { | |
345 | ipoib_warn(priv, "%s: Failed to query QP ret: %d\n", | |
346 | __func__, ret); | |
347 | goto free_res; | |
348 | } | |
349 | pr_info("%s: QP: 0x%x is in state: %d\n", | |
350 | __func__, priv->qp->qp_num, qp_attr.qp_state); | |
351 | ||
352 | /* currently support only in SQE->RTS transition*/ | |
353 | if (qp_attr.qp_state == IB_QPS_SQE) { | |
354 | qp_attr.qp_state = IB_QPS_RTS; | |
355 | ||
356 | ret = ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE); | |
357 | if (ret) { | |
358 | pr_warn("failed(%d) modify QP:0x%x SQE->RTS\n", | |
359 | ret, priv->qp->qp_num); | |
360 | goto free_res; | |
361 | } | |
362 | pr_info("%s: QP: 0x%x moved from IB_QPS_SQE to IB_QPS_RTS\n", | |
363 | __func__, priv->qp->qp_num); | |
364 | } else { | |
365 | pr_warn("QP (%d) will stay in state: %d\n", | |
366 | priv->qp->qp_num, qp_attr.qp_state); | |
367 | } | |
368 | ||
369 | free_res: | |
370 | kfree(qp_work); | |
371 | } | |
372 | ||
2439a6e6 RD |
373 | static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) |
374 | { | |
375 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
376 | unsigned int wr_id = wc->wr_id; | |
377 | struct ipoib_tx_buf *tx_req; | |
1da177e4 | 378 | |
a89875fc RD |
379 | ipoib_dbg_data(priv, "send completion: id %d, status: %d\n", |
380 | wr_id, wc->status); | |
1da177e4 | 381 | |
2439a6e6 RD |
382 | if (unlikely(wr_id >= ipoib_sendq_size)) { |
383 | ipoib_warn(priv, "send completion event with wrid %d (> %d)\n", | |
384 | wr_id, ipoib_sendq_size); | |
385 | return; | |
1da177e4 | 386 | } |
2439a6e6 RD |
387 | |
388 | tx_req = &priv->tx_ring[wr_id]; | |
389 | ||
c4268778 | 390 | ipoib_dma_unmap_tx(priv, tx_req); |
2439a6e6 | 391 | |
de903512 RD |
392 | ++dev->stats.tx_packets; |
393 | dev->stats.tx_bytes += tx_req->skb->len; | |
2439a6e6 RD |
394 | |
395 | dev_kfree_skb_any(tx_req->skb); | |
396 | ||
2439a6e6 | 397 | ++priv->tx_tail; |
1b524963 MT |
398 | if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) && |
399 | netif_queue_stopped(dev) && | |
400 | test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) | |
2439a6e6 | 401 | netif_wake_queue(dev); |
2439a6e6 RD |
402 | |
403 | if (wc->status != IB_WC_SUCCESS && | |
2c010730 ES |
404 | wc->status != IB_WC_WR_FLUSH_ERR) { |
405 | struct ipoib_qp_state_validate *qp_work; | |
2439a6e6 RD |
406 | ipoib_warn(priv, "failed send event " |
407 | "(status=%d, wrid=%d vend_err %x)\n", | |
408 | wc->status, wr_id, wc->vendor_err); | |
2c010730 ES |
409 | qp_work = kzalloc(sizeof(*qp_work), GFP_ATOMIC); |
410 | if (!qp_work) { | |
411 | ipoib_warn(priv, "%s Failed alloc ipoib_qp_state_validate for qp: 0x%x\n", | |
412 | __func__, priv->qp->qp_num); | |
413 | return; | |
414 | } | |
415 | ||
416 | INIT_WORK(&qp_work->work, ipoib_qp_state_validate_work); | |
417 | qp_work->priv = priv; | |
418 | queue_work(priv->wq, &qp_work->work); | |
419 | } | |
2439a6e6 RD |
420 | } |
421 | ||
f56bcd80 EC |
422 | static int poll_tx(struct ipoib_dev_priv *priv) |
423 | { | |
424 | int n, i; | |
425 | ||
426 | n = ib_poll_cq(priv->send_cq, MAX_SEND_CQE, priv->send_wc); | |
427 | for (i = 0; i < n; ++i) | |
428 | ipoib_ib_handle_tx_wc(priv->dev, priv->send_wc + i); | |
429 | ||
430 | return n == MAX_SEND_CQE; | |
431 | } | |
432 | ||
bea3348e | 433 | int ipoib_poll(struct napi_struct *napi, int budget) |
2439a6e6 | 434 | { |
bea3348e SH |
435 | struct ipoib_dev_priv *priv = container_of(napi, struct ipoib_dev_priv, napi); |
436 | struct net_device *dev = priv->dev; | |
8d1cc86a RD |
437 | int done; |
438 | int t; | |
8d1cc86a RD |
439 | int n, i; |
440 | ||
441 | done = 0; | |
8d1cc86a | 442 | |
bea3348e SH |
443 | poll_more: |
444 | while (done < budget) { | |
445 | int max = (budget - done); | |
446 | ||
8d1cc86a | 447 | t = min(IPOIB_NUM_WC, max); |
f56bcd80 | 448 | n = ib_poll_cq(priv->recv_cq, t, priv->ibwc); |
8d1cc86a | 449 | |
bea3348e | 450 | for (i = 0; i < n; i++) { |
8d1cc86a RD |
451 | struct ib_wc *wc = priv->ibwc + i; |
452 | ||
1b524963 | 453 | if (wc->wr_id & IPOIB_OP_RECV) { |
8d1cc86a | 454 | ++done; |
1b524963 MT |
455 | if (wc->wr_id & IPOIB_OP_CM) |
456 | ipoib_cm_handle_rx_wc(dev, wc); | |
457 | else | |
458 | ipoib_ib_handle_rx_wc(dev, wc); | |
f56bcd80 EC |
459 | } else |
460 | ipoib_cm_handle_tx_wc(priv->dev, wc); | |
8d1cc86a RD |
461 | } |
462 | ||
bea3348e | 463 | if (n != t) |
8d1cc86a | 464 | break; |
8d1cc86a RD |
465 | } |
466 | ||
bea3348e | 467 | if (done < budget) { |
288379f0 | 468 | napi_complete(napi); |
f56bcd80 | 469 | if (unlikely(ib_req_notify_cq(priv->recv_cq, |
8d1cc86a RD |
470 | IB_CQ_NEXT_COMP | |
471 | IB_CQ_REPORT_MISSED_EVENTS)) && | |
288379f0 | 472 | napi_reschedule(napi)) |
bea3348e | 473 | goto poll_more; |
8d1cc86a RD |
474 | } |
475 | ||
bea3348e | 476 | return done; |
1da177e4 LT |
477 | } |
478 | ||
479 | void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr) | |
480 | { | |
bea3348e SH |
481 | struct net_device *dev = dev_ptr; |
482 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
483 | ||
288379f0 | 484 | napi_schedule(&priv->napi); |
1da177e4 LT |
485 | } |
486 | ||
57ce41d1 EC |
487 | static void drain_tx_cq(struct net_device *dev) |
488 | { | |
489 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
57ce41d1 | 490 | |
943c246e | 491 | netif_tx_lock(dev); |
57ce41d1 EC |
492 | while (poll_tx(priv)) |
493 | ; /* nothing */ | |
494 | ||
495 | if (netif_queue_stopped(dev)) | |
496 | mod_timer(&priv->poll_timer, jiffies + 1); | |
497 | ||
943c246e | 498 | netif_tx_unlock(dev); |
57ce41d1 EC |
499 | } |
500 | ||
501 | void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr) | |
502 | { | |
943c246e RD |
503 | struct ipoib_dev_priv *priv = netdev_priv(dev_ptr); |
504 | ||
505 | mod_timer(&priv->poll_timer, jiffies); | |
57ce41d1 EC |
506 | } |
507 | ||
1da177e4 LT |
508 | static inline int post_send(struct ipoib_dev_priv *priv, |
509 | unsigned int wr_id, | |
510 | struct ib_ah *address, u32 qpn, | |
40ca1988 EC |
511 | struct ipoib_tx_buf *tx_req, |
512 | void *head, int hlen) | |
1da177e4 LT |
513 | { |
514 | struct ib_send_wr *bad_wr; | |
40ca1988 | 515 | struct sk_buff *skb = tx_req->skb; |
40ca1988 | 516 | |
c4268778 | 517 | ipoib_build_sge(priv, tx_req); |
1da177e4 | 518 | |
e622f2f4 CH |
519 | priv->tx_wr.wr.wr_id = wr_id; |
520 | priv->tx_wr.remote_qpn = qpn; | |
521 | priv->tx_wr.ah = address; | |
1da177e4 | 522 | |
40ca1988 | 523 | if (head) { |
e622f2f4 CH |
524 | priv->tx_wr.mss = skb_shinfo(skb)->gso_size; |
525 | priv->tx_wr.header = head; | |
526 | priv->tx_wr.hlen = hlen; | |
527 | priv->tx_wr.wr.opcode = IB_WR_LSO; | |
40ca1988 | 528 | } else |
e622f2f4 | 529 | priv->tx_wr.wr.opcode = IB_WR_SEND; |
40ca1988 | 530 | |
e622f2f4 | 531 | return ib_post_send(priv->qp, &priv->tx_wr.wr, &bad_wr); |
1da177e4 LT |
532 | } |
533 | ||
534 | void ipoib_send(struct net_device *dev, struct sk_buff *skb, | |
535 | struct ipoib_ah *address, u32 qpn) | |
536 | { | |
537 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1993d683 | 538 | struct ipoib_tx_buf *tx_req; |
a48f509b | 539 | int hlen, rc; |
40ca1988 | 540 | void *phead; |
78a50a5e | 541 | unsigned usable_sge = priv->max_send_sge - !!skb_headlen(skb); |
40ca1988 EC |
542 | |
543 | if (skb_is_gso(skb)) { | |
544 | hlen = skb_transport_offset(skb) + tcp_hdrlen(skb); | |
545 | phead = skb->data; | |
546 | if (unlikely(!skb_pull(skb, hlen))) { | |
547 | ipoib_warn(priv, "linear data too small\n"); | |
548 | ++dev->stats.tx_dropped; | |
549 | ++dev->stats.tx_errors; | |
550 | dev_kfree_skb_any(skb); | |
551 | return; | |
552 | } | |
553 | } else { | |
554 | if (unlikely(skb->len > priv->mcast_mtu + IPOIB_ENCAP_LEN)) { | |
555 | ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n", | |
556 | skb->len, priv->mcast_mtu + IPOIB_ENCAP_LEN); | |
557 | ++dev->stats.tx_dropped; | |
558 | ++dev->stats.tx_errors; | |
559 | ipoib_cm_skb_too_long(dev, skb, priv->mcast_mtu); | |
560 | return; | |
561 | } | |
562 | phead = NULL; | |
563 | hlen = 0; | |
1da177e4 | 564 | } |
78a50a5e HWR |
565 | if (skb_shinfo(skb)->nr_frags > usable_sge) { |
566 | if (skb_linearize(skb) < 0) { | |
567 | ipoib_warn(priv, "skb could not be linearized\n"); | |
568 | ++dev->stats.tx_dropped; | |
569 | ++dev->stats.tx_errors; | |
570 | dev_kfree_skb_any(skb); | |
571 | return; | |
572 | } | |
573 | /* Does skb_linearize return ok without reducing nr_frags? */ | |
574 | if (skb_shinfo(skb)->nr_frags > usable_sge) { | |
575 | ipoib_warn(priv, "too many frags after skb linearize\n"); | |
576 | ++dev->stats.tx_dropped; | |
577 | ++dev->stats.tx_errors; | |
578 | dev_kfree_skb_any(skb); | |
579 | return; | |
580 | } | |
581 | } | |
1da177e4 LT |
582 | |
583 | ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n", | |
584 | skb->len, address, qpn); | |
585 | ||
586 | /* | |
587 | * We put the skb into the tx_ring _before_ we call post_send() | |
588 | * because it's entirely possible that the completion handler will | |
589 | * run before we execute anything after the post_send(). That | |
590 | * means we have to make sure everything is properly recorded and | |
591 | * our state is consistent before we call post_send(). | |
592 | */ | |
0f485251 | 593 | tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)]; |
1da177e4 | 594 | tx_req->skb = skb; |
7143740d | 595 | if (unlikely(ipoib_dma_map_tx(priv->ca, tx_req))) { |
de903512 | 596 | ++dev->stats.tx_errors; |
73fbe8be RD |
597 | dev_kfree_skb_any(skb); |
598 | return; | |
599 | } | |
1da177e4 | 600 | |
6046136c | 601 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
e622f2f4 | 602 | priv->tx_wr.wr.send_flags |= IB_SEND_IP_CSUM; |
6046136c | 603 | else |
e622f2f4 | 604 | priv->tx_wr.wr.send_flags &= ~IB_SEND_IP_CSUM; |
6046136c | 605 | |
57ce41d1 EC |
606 | if (++priv->tx_outstanding == ipoib_sendq_size) { |
607 | ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n"); | |
608 | if (ib_req_notify_cq(priv->send_cq, IB_CQ_NEXT_COMP)) | |
609 | ipoib_warn(priv, "request notify on send CQ failed\n"); | |
610 | netif_stop_queue(dev); | |
611 | } | |
612 | ||
7e5a90c2 SP |
613 | skb_orphan(skb); |
614 | skb_dst_drop(skb); | |
615 | ||
a48f509b OG |
616 | rc = post_send(priv, priv->tx_head & (ipoib_sendq_size - 1), |
617 | address->ah, qpn, tx_req, phead, hlen); | |
618 | if (unlikely(rc)) { | |
619 | ipoib_warn(priv, "post_send failed, error %d\n", rc); | |
de903512 | 620 | ++dev->stats.tx_errors; |
57ce41d1 | 621 | --priv->tx_outstanding; |
c4268778 | 622 | ipoib_dma_unmap_tx(priv, tx_req); |
1da177e4 | 623 | dev_kfree_skb_any(skb); |
57ce41d1 EC |
624 | if (netif_queue_stopped(dev)) |
625 | netif_wake_queue(dev); | |
1da177e4 LT |
626 | } else { |
627 | dev->trans_start = jiffies; | |
628 | ||
629 | address->last_send = priv->tx_head; | |
630 | ++priv->tx_head; | |
1da177e4 | 631 | } |
f56bcd80 EC |
632 | |
633 | if (unlikely(priv->tx_outstanding > MAX_SEND_CQE)) | |
57ce41d1 EC |
634 | while (poll_tx(priv)) |
635 | ; /* nothing */ | |
1da177e4 LT |
636 | } |
637 | ||
638 | static void __ipoib_reap_ah(struct net_device *dev) | |
639 | { | |
640 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
641 | struct ipoib_ah *ah, *tah; | |
642 | LIST_HEAD(remove_list); | |
943c246e RD |
643 | unsigned long flags; |
644 | ||
645 | netif_tx_lock_bh(dev); | |
646 | spin_lock_irqsave(&priv->lock, flags); | |
1da177e4 | 647 | |
1da177e4 | 648 | list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list) |
2181858b | 649 | if ((int) priv->tx_tail - (int) ah->last_send >= 0) { |
1da177e4 | 650 | list_del(&ah->list); |
31c02e21 RD |
651 | ib_destroy_ah(ah->ah); |
652 | kfree(ah); | |
1da177e4 | 653 | } |
943c246e RD |
654 | |
655 | spin_unlock_irqrestore(&priv->lock, flags); | |
656 | netif_tx_unlock_bh(dev); | |
1da177e4 LT |
657 | } |
658 | ||
c4028958 | 659 | void ipoib_reap_ah(struct work_struct *work) |
1da177e4 | 660 | { |
c4028958 DH |
661 | struct ipoib_dev_priv *priv = |
662 | container_of(work, struct ipoib_dev_priv, ah_reap_task.work); | |
663 | struct net_device *dev = priv->dev; | |
1da177e4 LT |
664 | |
665 | __ipoib_reap_ah(dev); | |
666 | ||
667 | if (!test_bit(IPOIB_STOP_REAPER, &priv->flags)) | |
0b39578b | 668 | queue_delayed_work(priv->wq, &priv->ah_reap_task, |
69fc507a | 669 | round_jiffies_relative(HZ)); |
1da177e4 LT |
670 | } |
671 | ||
efc82eee | 672 | static void ipoib_flush_ah(struct net_device *dev) |
e135106f DL |
673 | { |
674 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
675 | ||
676 | cancel_delayed_work(&priv->ah_reap_task); | |
efc82eee | 677 | flush_workqueue(priv->wq); |
e135106f DL |
678 | ipoib_reap_ah(&priv->ah_reap_task.work); |
679 | } | |
680 | ||
efc82eee | 681 | static void ipoib_stop_ah(struct net_device *dev) |
e135106f DL |
682 | { |
683 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
684 | ||
685 | set_bit(IPOIB_STOP_REAPER, &priv->flags); | |
efc82eee | 686 | ipoib_flush_ah(dev); |
e135106f DL |
687 | } |
688 | ||
57ce41d1 EC |
689 | static void ipoib_ib_tx_timer_func(unsigned long ctx) |
690 | { | |
691 | drain_tx_cq((struct net_device *)ctx); | |
692 | } | |
693 | ||
efc82eee | 694 | int ipoib_ib_dev_open(struct net_device *dev) |
1da177e4 LT |
695 | { |
696 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
697 | int ret; | |
698 | ||
dd57c930 AE |
699 | ipoib_pkey_dev_check_presence(dev); |
700 | ||
701 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
702 | ipoib_warn(priv, "P_Key 0x%04x is %s\n", priv->pkey, | |
703 | (!(priv->pkey & 0x7fff) ? "Invalid" : "not found")); | |
26bbf13c YE |
704 | return -1; |
705 | } | |
26bbf13c | 706 | |
5b6810e0 | 707 | ret = ipoib_init_qp(dev); |
1da177e4 | 708 | if (ret) { |
5b6810e0 | 709 | ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret); |
1da177e4 LT |
710 | return -1; |
711 | } | |
712 | ||
713 | ret = ipoib_ib_post_receives(dev); | |
714 | if (ret) { | |
715 | ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret); | |
c2bb5628 | 716 | goto dev_stop; |
1da177e4 LT |
717 | } |
718 | ||
839fcaba MT |
719 | ret = ipoib_cm_dev_open(dev); |
720 | if (ret) { | |
24bd1e4e | 721 | ipoib_warn(priv, "ipoib_cm_dev_open returned %d\n", ret); |
c2bb5628 | 722 | goto dev_stop; |
839fcaba MT |
723 | } |
724 | ||
1da177e4 | 725 | clear_bit(IPOIB_STOP_REAPER, &priv->flags); |
0b39578b | 726 | queue_delayed_work(priv->wq, &priv->ah_reap_task, |
69fc507a | 727 | round_jiffies_relative(HZ)); |
1da177e4 | 728 | |
e028cc55 YE |
729 | if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
730 | napi_enable(&priv->napi); | |
7a343d4c | 731 | |
1da177e4 | 732 | return 0; |
c2bb5628 ES |
733 | dev_stop: |
734 | if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) | |
735 | napi_enable(&priv->napi); | |
efc82eee | 736 | ipoib_ib_dev_stop(dev); |
c2bb5628 | 737 | return -1; |
1da177e4 LT |
738 | } |
739 | ||
db84f880 | 740 | void ipoib_pkey_dev_check_presence(struct net_device *dev) |
7a343d4c LA |
741 | { |
742 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
7a343d4c | 743 | |
dd57c930 AE |
744 | if (!(priv->pkey & 0x7fff) || |
745 | ib_find_pkey(priv->ca, priv->port, priv->pkey, | |
746 | &priv->pkey_index)) | |
7a343d4c LA |
747 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); |
748 | else | |
749 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
750 | } | |
751 | ||
1da177e4 LT |
752 | int ipoib_ib_dev_up(struct net_device *dev) |
753 | { | |
754 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
755 | ||
7a343d4c LA |
756 | ipoib_pkey_dev_check_presence(dev); |
757 | ||
758 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
759 | ipoib_dbg(priv, "PKEY is not assigned.\n"); | |
760 | return 0; | |
761 | } | |
762 | ||
1da177e4 LT |
763 | set_bit(IPOIB_FLAG_OPER_UP, &priv->flags); |
764 | ||
765 | return ipoib_mcast_start_thread(dev); | |
766 | } | |
767 | ||
efc82eee | 768 | int ipoib_ib_dev_down(struct net_device *dev) |
1da177e4 LT |
769 | { |
770 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
771 | ||
772 | ipoib_dbg(priv, "downing ib_dev\n"); | |
773 | ||
774 | clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags); | |
775 | netif_carrier_off(dev); | |
776 | ||
efc82eee | 777 | ipoib_mcast_stop_thread(dev); |
1da177e4 LT |
778 | ipoib_mcast_dev_flush(dev); |
779 | ||
1da177e4 LT |
780 | ipoib_flush_paths(dev); |
781 | ||
782 | return 0; | |
783 | } | |
784 | ||
785 | static int recvs_pending(struct net_device *dev) | |
786 | { | |
787 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
788 | int pending = 0; | |
789 | int i; | |
790 | ||
0f485251 | 791 | for (i = 0; i < ipoib_recvq_size; ++i) |
1da177e4 LT |
792 | if (priv->rx_ring[i].skb) |
793 | ++pending; | |
794 | ||
795 | return pending; | |
796 | } | |
797 | ||
2dfbfc37 MT |
798 | void ipoib_drain_cq(struct net_device *dev) |
799 | { | |
800 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
801 | int i, n; | |
943c246e RD |
802 | |
803 | /* | |
804 | * We call completion handling routines that expect to be | |
805 | * called from the BH-disabled NAPI poll context, so disable | |
806 | * BHs here too. | |
807 | */ | |
808 | local_bh_disable(); | |
809 | ||
2dfbfc37 | 810 | do { |
f56bcd80 | 811 | n = ib_poll_cq(priv->recv_cq, IPOIB_NUM_WC, priv->ibwc); |
2dfbfc37 | 812 | for (i = 0; i < n; ++i) { |
ce423ef5 RD |
813 | /* |
814 | * Convert any successful completions to flush | |
815 | * errors to avoid passing packets up the | |
816 | * stack after bringing the device down. | |
817 | */ | |
818 | if (priv->ibwc[i].status == IB_WC_SUCCESS) | |
819 | priv->ibwc[i].status = IB_WC_WR_FLUSH_ERR; | |
820 | ||
1b524963 MT |
821 | if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) { |
822 | if (priv->ibwc[i].wr_id & IPOIB_OP_CM) | |
823 | ipoib_cm_handle_rx_wc(dev, priv->ibwc + i); | |
824 | else | |
825 | ipoib_ib_handle_rx_wc(dev, priv->ibwc + i); | |
f56bcd80 EC |
826 | } else |
827 | ipoib_cm_handle_tx_wc(dev, priv->ibwc + i); | |
2dfbfc37 MT |
828 | } |
829 | } while (n == IPOIB_NUM_WC); | |
f56bcd80 EC |
830 | |
831 | while (poll_tx(priv)) | |
832 | ; /* nothing */ | |
943c246e RD |
833 | |
834 | local_bh_enable(); | |
2dfbfc37 MT |
835 | } |
836 | ||
efc82eee | 837 | int ipoib_ib_dev_stop(struct net_device *dev) |
1da177e4 LT |
838 | { |
839 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
840 | struct ib_qp_attr qp_attr; | |
1da177e4 | 841 | unsigned long begin; |
1993d683 | 842 | struct ipoib_tx_buf *tx_req; |
2dfbfc37 | 843 | int i; |
1da177e4 | 844 | |
e028cc55 YE |
845 | if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
846 | napi_disable(&priv->napi); | |
7a343d4c | 847 | |
839fcaba MT |
848 | ipoib_cm_dev_stop(dev); |
849 | ||
3bc12e75 RD |
850 | /* |
851 | * Move our QP to the error state and then reinitialize in | |
852 | * when all work requests have completed or have been flushed. | |
853 | */ | |
1da177e4 | 854 | qp_attr.qp_state = IB_QPS_ERR; |
3bc12e75 | 855 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
856 | ipoib_warn(priv, "Failed to modify QP to ERROR state\n"); |
857 | ||
858 | /* Wait for all sends and receives to complete */ | |
859 | begin = jiffies; | |
860 | ||
861 | while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) { | |
862 | if (time_after(jiffies, begin + 5 * HZ)) { | |
863 | ipoib_warn(priv, "timing out; %d sends %d receives not completed\n", | |
864 | priv->tx_head - priv->tx_tail, recvs_pending(dev)); | |
865 | ||
866 | /* | |
867 | * assume the HW is wedged and just free up | |
868 | * all our pending work requests. | |
869 | */ | |
2181858b | 870 | while ((int) priv->tx_tail - (int) priv->tx_head < 0) { |
1da177e4 | 871 | tx_req = &priv->tx_ring[priv->tx_tail & |
0f485251 | 872 | (ipoib_sendq_size - 1)]; |
c4268778 | 873 | ipoib_dma_unmap_tx(priv, tx_req); |
1da177e4 LT |
874 | dev_kfree_skb_any(tx_req->skb); |
875 | ++priv->tx_tail; | |
1b524963 | 876 | --priv->tx_outstanding; |
1da177e4 LT |
877 | } |
878 | ||
37ccf9df RC |
879 | for (i = 0; i < ipoib_recvq_size; ++i) { |
880 | struct ipoib_rx_buf *rx_req; | |
881 | ||
882 | rx_req = &priv->rx_ring[i]; | |
883 | if (!rx_req->skb) | |
884 | continue; | |
bc7b3a36 SM |
885 | ipoib_ud_dma_unmap_rx(priv, |
886 | priv->rx_ring[i].mapping); | |
37ccf9df RC |
887 | dev_kfree_skb_any(rx_req->skb); |
888 | rx_req->skb = NULL; | |
889 | } | |
1da177e4 LT |
890 | |
891 | goto timeout; | |
892 | } | |
893 | ||
2dfbfc37 | 894 | ipoib_drain_cq(dev); |
8d1cc86a | 895 | |
1da177e4 LT |
896 | msleep(1); |
897 | } | |
898 | ||
899 | ipoib_dbg(priv, "All sends and receives done.\n"); | |
900 | ||
901 | timeout: | |
57ce41d1 | 902 | del_timer_sync(&priv->poll_timer); |
1da177e4 | 903 | qp_attr.qp_state = IB_QPS_RESET; |
3bc12e75 | 904 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
905 | ipoib_warn(priv, "Failed to modify QP to RESET state\n"); |
906 | ||
efc82eee | 907 | ipoib_flush_ah(dev); |
1da177e4 | 908 | |
f56bcd80 | 909 | ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP); |
8d1cc86a | 910 | |
1da177e4 LT |
911 | return 0; |
912 | } | |
913 | ||
914 | int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port) | |
915 | { | |
916 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
917 | ||
918 | priv->ca = ca; | |
919 | priv->port = port; | |
920 | priv->qp = NULL; | |
921 | ||
922 | if (ipoib_transport_dev_init(dev, ca)) { | |
923 | printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name); | |
924 | return -ENODEV; | |
925 | } | |
926 | ||
2767840a RD |
927 | setup_timer(&priv->poll_timer, ipoib_ib_tx_timer_func, |
928 | (unsigned long) dev); | |
929 | ||
1da177e4 | 930 | if (dev->flags & IFF_UP) { |
efc82eee | 931 | if (ipoib_ib_dev_open(dev)) { |
1da177e4 LT |
932 | ipoib_transport_dev_cleanup(dev); |
933 | return -ENODEV; | |
934 | } | |
935 | } | |
936 | ||
937 | return 0; | |
938 | } | |
939 | ||
c2904141 ES |
940 | /* |
941 | * Takes whatever value which is in pkey index 0 and updates priv->pkey | |
942 | * returns 0 if the pkey value was changed. | |
943 | */ | |
944 | static inline int update_parent_pkey(struct ipoib_dev_priv *priv) | |
945 | { | |
946 | int result; | |
947 | u16 prev_pkey; | |
948 | ||
949 | prev_pkey = priv->pkey; | |
950 | result = ib_query_pkey(priv->ca, priv->port, 0, &priv->pkey); | |
951 | if (result) { | |
952 | ipoib_warn(priv, "ib_query_pkey port %d failed (ret = %d)\n", | |
953 | priv->port, result); | |
954 | return result; | |
955 | } | |
956 | ||
957 | priv->pkey |= 0x8000; | |
958 | ||
959 | if (prev_pkey != priv->pkey) { | |
960 | ipoib_dbg(priv, "pkey changed from 0x%x to 0x%x\n", | |
961 | prev_pkey, priv->pkey); | |
962 | /* | |
963 | * Update the pkey in the broadcast address, while making sure to set | |
964 | * the full membership bit, so that we join the right broadcast group. | |
965 | */ | |
966 | priv->dev->broadcast[8] = priv->pkey >> 8; | |
967 | priv->dev->broadcast[9] = priv->pkey & 0xff; | |
968 | return 0; | |
969 | } | |
970 | ||
971 | return 1; | |
972 | } | |
dd57c930 AE |
973 | /* |
974 | * returns 0 if pkey value was found in a different slot. | |
975 | */ | |
976 | static inline int update_child_pkey(struct ipoib_dev_priv *priv) | |
977 | { | |
978 | u16 old_index = priv->pkey_index; | |
979 | ||
980 | priv->pkey_index = 0; | |
981 | ipoib_pkey_dev_check_presence(priv->dev); | |
982 | ||
983 | if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags) && | |
984 | (old_index == priv->pkey_index)) | |
985 | return 1; | |
986 | return 0; | |
987 | } | |
c2904141 | 988 | |
ee1e2c82 | 989 | static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, |
8b7cce0d HE |
990 | enum ipoib_flush_level level, |
991 | int nesting) | |
1da177e4 | 992 | { |
26bbf13c | 993 | struct ipoib_dev_priv *cpriv; |
c4028958 | 994 | struct net_device *dev = priv->dev; |
c2904141 | 995 | int result; |
26bbf13c | 996 | |
8b7cce0d | 997 | down_read_nested(&priv->vlan_rwsem, nesting); |
1da177e4 | 998 | |
26bbf13c YE |
999 | /* |
1000 | * Flush any child interfaces too -- they might be up even if | |
1001 | * the parent is down. | |
1002 | */ | |
1003 | list_for_each_entry(cpriv, &priv->child_intfs, list) | |
8b7cce0d | 1004 | __ipoib_ib_dev_flush(cpriv, level, nesting + 1); |
26bbf13c | 1005 | |
f47944cc | 1006 | up_read(&priv->vlan_rwsem); |
26bbf13c | 1007 | |
dd57c930 AE |
1008 | if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags) && |
1009 | level != IPOIB_FLUSH_HEAVY) { | |
7a343d4c LA |
1010 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n"); |
1011 | return; | |
1012 | } | |
1013 | ||
1014 | if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { | |
dd57c930 AE |
1015 | /* interface is down. update pkey and leave. */ |
1016 | if (level == IPOIB_FLUSH_HEAVY) { | |
1017 | if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) | |
1018 | update_parent_pkey(priv); | |
1019 | else | |
1020 | update_child_pkey(priv); | |
1021 | } | |
7a343d4c | 1022 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n"); |
1da177e4 | 1023 | return; |
7a343d4c | 1024 | } |
1da177e4 | 1025 | |
ee1e2c82 | 1026 | if (level == IPOIB_FLUSH_HEAVY) { |
c2904141 ES |
1027 | /* child devices chase their origin pkey value, while non-child |
1028 | * (parent) devices should always takes what present in pkey index 0 | |
1029 | */ | |
1030 | if (test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { | |
dd57c930 AE |
1031 | result = update_child_pkey(priv); |
1032 | if (result) { | |
1033 | /* restart QP only if P_Key index is changed */ | |
c2904141 | 1034 | ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n"); |
9fdd5e5b | 1035 | return; |
c2904141 | 1036 | } |
dd57c930 | 1037 | |
c2904141 ES |
1038 | } else { |
1039 | result = update_parent_pkey(priv); | |
1040 | /* restart QP only if P_Key value changed */ | |
1041 | if (result) { | |
1042 | ipoib_dbg(priv, "Not flushing - P_Key value not changed.\n"); | |
1043 | return; | |
1044 | } | |
26bbf13c | 1045 | } |
26bbf13c YE |
1046 | } |
1047 | ||
ee1e2c82 MS |
1048 | if (level == IPOIB_FLUSH_LIGHT) { |
1049 | ipoib_mark_paths_invalid(dev); | |
1050 | ipoib_mcast_dev_flush(dev); | |
efc82eee | 1051 | ipoib_flush_ah(dev); |
ee1e2c82 | 1052 | } |
1da177e4 | 1053 | |
ee1e2c82 | 1054 | if (level >= IPOIB_FLUSH_NORMAL) |
efc82eee | 1055 | ipoib_ib_dev_down(dev); |
1da177e4 | 1056 | |
ee1e2c82 | 1057 | if (level == IPOIB_FLUSH_HEAVY) { |
dd57c930 | 1058 | if (test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) |
efc82eee DL |
1059 | ipoib_ib_dev_stop(dev); |
1060 | if (ipoib_ib_dev_open(dev) != 0) | |
dd57c930 AE |
1061 | return; |
1062 | if (netif_queue_stopped(dev)) | |
1063 | netif_start_queue(dev); | |
26bbf13c YE |
1064 | } |
1065 | ||
1da177e4 LT |
1066 | /* |
1067 | * The device could have been brought down between the start and when | |
1068 | * we get here, don't bring it back up if it's not configured up | |
1069 | */ | |
5ccd0255 | 1070 | if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { |
ee1e2c82 MS |
1071 | if (level >= IPOIB_FLUSH_NORMAL) |
1072 | ipoib_ib_dev_up(dev); | |
c4028958 | 1073 | ipoib_mcast_restart_task(&priv->restart_task); |
5ccd0255 | 1074 | } |
26bbf13c | 1075 | } |
1da177e4 | 1076 | |
ee1e2c82 MS |
1077 | void ipoib_ib_dev_flush_light(struct work_struct *work) |
1078 | { | |
1079 | struct ipoib_dev_priv *priv = | |
1080 | container_of(work, struct ipoib_dev_priv, flush_light); | |
1081 | ||
8b7cce0d | 1082 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_LIGHT, 0); |
ee1e2c82 MS |
1083 | } |
1084 | ||
1085 | void ipoib_ib_dev_flush_normal(struct work_struct *work) | |
26bbf13c YE |
1086 | { |
1087 | struct ipoib_dev_priv *priv = | |
ee1e2c82 | 1088 | container_of(work, struct ipoib_dev_priv, flush_normal); |
4f71055a | 1089 | |
8b7cce0d | 1090 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_NORMAL, 0); |
26bbf13c | 1091 | } |
4f71055a | 1092 | |
ee1e2c82 | 1093 | void ipoib_ib_dev_flush_heavy(struct work_struct *work) |
26bbf13c YE |
1094 | { |
1095 | struct ipoib_dev_priv *priv = | |
ee1e2c82 | 1096 | container_of(work, struct ipoib_dev_priv, flush_heavy); |
26bbf13c | 1097 | |
8b7cce0d | 1098 | __ipoib_ib_dev_flush(priv, IPOIB_FLUSH_HEAVY, 0); |
1da177e4 LT |
1099 | } |
1100 | ||
1101 | void ipoib_ib_dev_cleanup(struct net_device *dev) | |
1102 | { | |
1103 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1104 | ||
1105 | ipoib_dbg(priv, "cleaning up ib_dev\n"); | |
a39c52ab ES |
1106 | /* |
1107 | * We must make sure there are no more (path) completions | |
1108 | * that may wish to touch priv fields that are no longer valid | |
1109 | */ | |
1110 | ipoib_flush_paths(dev); | |
1da177e4 | 1111 | |
efc82eee | 1112 | ipoib_mcast_stop_thread(dev); |
988bd503 | 1113 | ipoib_mcast_dev_flush(dev); |
1da177e4 | 1114 | |
e135106f DL |
1115 | /* |
1116 | * All of our ah references aren't free until after | |
1117 | * ipoib_mcast_dev_flush(), ipoib_flush_paths, and | |
1118 | * the neighbor garbage collection is stopped and reaped. | |
1119 | * That should all be done now, so make a final ah flush. | |
1120 | */ | |
efc82eee | 1121 | ipoib_stop_ah(dev); |
e135106f | 1122 | |
1da177e4 LT |
1123 | ipoib_transport_dev_cleanup(dev); |
1124 | } | |
1125 | ||
1da177e4 | 1126 |