Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. | |
2a1d9b7f RD |
3 | * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. |
4 | * Copyright (c) 2005 Mellanox Technologies. All rights reserved. | |
5 | * Copyright (c) 2004, 2005 Voltaire, Inc. All rights reserved. | |
1da177e4 LT |
6 | * |
7 | * This software is available to you under a choice of one of two | |
8 | * licenses. You may choose to be licensed under the terms of the GNU | |
9 | * General Public License (GPL) Version 2, available from the file | |
10 | * COPYING in the main directory of this source tree, or the | |
11 | * OpenIB.org BSD license below: | |
12 | * | |
13 | * Redistribution and use in source and binary forms, with or | |
14 | * without modification, are permitted provided that the following | |
15 | * conditions are met: | |
16 | * | |
17 | * - Redistributions of source code must retain the above | |
18 | * copyright notice, this list of conditions and the following | |
19 | * disclaimer. | |
20 | * | |
21 | * - Redistributions in binary form must reproduce the above | |
22 | * copyright notice, this list of conditions and the following | |
23 | * disclaimer in the documentation and/or other materials | |
24 | * provided with the distribution. | |
25 | * | |
26 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
27 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
28 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
29 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
30 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
31 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
32 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
33 | * SOFTWARE. | |
34 | * | |
35 | * $Id: ipoib_ib.c 1386 2004-12-27 16:23:17Z roland $ | |
36 | */ | |
37 | ||
38 | #include <linux/delay.h> | |
39 | #include <linux/dma-mapping.h> | |
40 | ||
a4d61e84 | 41 | #include <rdma/ib_cache.h> |
40ca1988 EC |
42 | #include <linux/ip.h> |
43 | #include <linux/tcp.h> | |
1da177e4 LT |
44 | |
45 | #include "ipoib.h" | |
46 | ||
47 | #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA | |
48 | static int data_debug_level; | |
49 | ||
50 | module_param(data_debug_level, int, 0644); | |
51 | MODULE_PARM_DESC(data_debug_level, | |
52 | "Enable data path debug tracing if > 0"); | |
53 | #endif | |
54 | ||
95ed644f | 55 | static DEFINE_MUTEX(pkey_mutex); |
1da177e4 LT |
56 | |
57 | struct ipoib_ah *ipoib_create_ah(struct net_device *dev, | |
58 | struct ib_pd *pd, struct ib_ah_attr *attr) | |
59 | { | |
60 | struct ipoib_ah *ah; | |
61 | ||
62 | ah = kmalloc(sizeof *ah, GFP_KERNEL); | |
63 | if (!ah) | |
64 | return NULL; | |
65 | ||
66 | ah->dev = dev; | |
67 | ah->last_send = 0; | |
68 | kref_init(&ah->ref); | |
69 | ||
70 | ah->ah = ib_create_ah(pd, attr); | |
71 | if (IS_ERR(ah->ah)) { | |
72 | kfree(ah); | |
73 | ah = NULL; | |
74 | } else | |
75 | ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah); | |
76 | ||
77 | return ah; | |
78 | } | |
79 | ||
80 | void ipoib_free_ah(struct kref *kref) | |
81 | { | |
82 | struct ipoib_ah *ah = container_of(kref, struct ipoib_ah, ref); | |
83 | struct ipoib_dev_priv *priv = netdev_priv(ah->dev); | |
84 | ||
85 | unsigned long flags; | |
86 | ||
31c02e21 RD |
87 | spin_lock_irqsave(&priv->lock, flags); |
88 | list_add_tail(&ah->list, &priv->dead_ahs); | |
89 | spin_unlock_irqrestore(&priv->lock, flags); | |
1da177e4 LT |
90 | } |
91 | ||
1993d683 | 92 | static int ipoib_ib_post_receive(struct net_device *dev, int id) |
1da177e4 | 93 | { |
1993d683 RD |
94 | struct ipoib_dev_priv *priv = netdev_priv(dev); |
95 | struct ib_sge list; | |
96 | struct ib_recv_wr param; | |
1da177e4 | 97 | struct ib_recv_wr *bad_wr; |
1993d683 RD |
98 | int ret; |
99 | ||
100 | list.addr = priv->rx_ring[id].mapping; | |
101 | list.length = IPOIB_BUF_SIZE; | |
102 | list.lkey = priv->mr->lkey; | |
103 | ||
104 | param.next = NULL; | |
105 | param.wr_id = id | IPOIB_OP_RECV; | |
106 | param.sg_list = &list; | |
107 | param.num_sge = 1; | |
108 | ||
109 | ret = ib_post_recv(priv->qp, ¶m, &bad_wr); | |
110 | if (unlikely(ret)) { | |
111 | ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret); | |
37ccf9df RC |
112 | ib_dma_unmap_single(priv->ca, priv->rx_ring[id].mapping, |
113 | IPOIB_BUF_SIZE, DMA_FROM_DEVICE); | |
1993d683 RD |
114 | dev_kfree_skb_any(priv->rx_ring[id].skb); |
115 | priv->rx_ring[id].skb = NULL; | |
116 | } | |
1da177e4 | 117 | |
1993d683 | 118 | return ret; |
1da177e4 LT |
119 | } |
120 | ||
1993d683 | 121 | static int ipoib_alloc_rx_skb(struct net_device *dev, int id) |
1da177e4 LT |
122 | { |
123 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
124 | struct sk_buff *skb; | |
37ccf9df | 125 | u64 addr; |
1da177e4 LT |
126 | |
127 | skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4); | |
1993d683 | 128 | if (!skb) |
1da177e4 | 129 | return -ENOMEM; |
1993d683 RD |
130 | |
131 | /* | |
132 | * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte | |
133 | * header. So we need 4 more bytes to get to 48 and align the | |
134 | * IP header to a multiple of 16. | |
135 | */ | |
136 | skb_reserve(skb, 4); | |
137 | ||
37ccf9df RC |
138 | addr = ib_dma_map_single(priv->ca, skb->data, IPOIB_BUF_SIZE, |
139 | DMA_FROM_DEVICE); | |
140 | if (unlikely(ib_dma_mapping_error(priv->ca, addr))) { | |
1da177e4 | 141 | dev_kfree_skb_any(skb); |
1993d683 | 142 | return -EIO; |
1da177e4 LT |
143 | } |
144 | ||
1993d683 RD |
145 | priv->rx_ring[id].skb = skb; |
146 | priv->rx_ring[id].mapping = addr; | |
147 | ||
148 | return 0; | |
1da177e4 LT |
149 | } |
150 | ||
151 | static int ipoib_ib_post_receives(struct net_device *dev) | |
152 | { | |
153 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
154 | int i; | |
155 | ||
0f485251 | 156 | for (i = 0; i < ipoib_recvq_size; ++i) { |
1993d683 RD |
157 | if (ipoib_alloc_rx_skb(dev, i)) { |
158 | ipoib_warn(priv, "failed to allocate receive buffer %d\n", i); | |
159 | return -ENOMEM; | |
160 | } | |
1da177e4 LT |
161 | if (ipoib_ib_post_receive(dev, i)) { |
162 | ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i); | |
163 | return -EIO; | |
164 | } | |
165 | } | |
166 | ||
167 | return 0; | |
168 | } | |
169 | ||
2439a6e6 | 170 | static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc) |
1da177e4 LT |
171 | { |
172 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
2439a6e6 RD |
173 | unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV; |
174 | struct sk_buff *skb; | |
37ccf9df | 175 | u64 addr; |
1da177e4 | 176 | |
a89875fc RD |
177 | ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n", |
178 | wr_id, wc->status); | |
1da177e4 | 179 | |
2439a6e6 RD |
180 | if (unlikely(wr_id >= ipoib_recvq_size)) { |
181 | ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n", | |
182 | wr_id, ipoib_recvq_size); | |
183 | return; | |
184 | } | |
185 | ||
186 | skb = priv->rx_ring[wr_id].skb; | |
187 | addr = priv->rx_ring[wr_id].mapping; | |
188 | ||
189 | if (unlikely(wc->status != IB_WC_SUCCESS)) { | |
190 | if (wc->status != IB_WC_WR_FLUSH_ERR) | |
191 | ipoib_warn(priv, "failed recv event " | |
192 | "(status=%d, wrid=%d vend_err %x)\n", | |
193 | wc->status, wr_id, wc->vendor_err); | |
37ccf9df RC |
194 | ib_dma_unmap_single(priv->ca, addr, |
195 | IPOIB_BUF_SIZE, DMA_FROM_DEVICE); | |
2439a6e6 RD |
196 | dev_kfree_skb_any(skb); |
197 | priv->rx_ring[wr_id].skb = NULL; | |
198 | return; | |
199 | } | |
1da177e4 | 200 | |
1b844afe RD |
201 | /* |
202 | * Drop packets that this interface sent, ie multicast packets | |
203 | * that the HCA has replicated. | |
204 | */ | |
205 | if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num) | |
206 | goto repost; | |
207 | ||
2439a6e6 RD |
208 | /* |
209 | * If we can't allocate a new RX buffer, dump | |
210 | * this packet and reuse the old buffer. | |
211 | */ | |
212 | if (unlikely(ipoib_alloc_rx_skb(dev, wr_id))) { | |
de903512 | 213 | ++dev->stats.rx_dropped; |
2439a6e6 RD |
214 | goto repost; |
215 | } | |
216 | ||
217 | ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n", | |
218 | wc->byte_len, wc->slid); | |
219 | ||
37ccf9df | 220 | ib_dma_unmap_single(priv->ca, addr, IPOIB_BUF_SIZE, DMA_FROM_DEVICE); |
2439a6e6 RD |
221 | |
222 | skb_put(skb, wc->byte_len); | |
223 | skb_pull(skb, IB_GRH_BYTES); | |
224 | ||
1b844afe RD |
225 | skb->protocol = ((struct ipoib_header *) skb->data)->proto; |
226 | skb_reset_mac_header(skb); | |
227 | skb_pull(skb, IPOIB_ENCAP_LEN); | |
228 | ||
229 | dev->last_rx = jiffies; | |
de903512 RD |
230 | ++dev->stats.rx_packets; |
231 | dev->stats.rx_bytes += skb->len; | |
1b844afe RD |
232 | |
233 | skb->dev = dev; | |
234 | /* XXX get correct PACKET_ type here */ | |
235 | skb->pkt_type = PACKET_HOST; | |
6046136c EC |
236 | |
237 | if (test_bit(IPOIB_FLAG_CSUM, &priv->flags) && likely(wc->csum_ok)) | |
238 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
239 | ||
1b844afe | 240 | netif_receive_skb(skb); |
1da177e4 | 241 | |
2439a6e6 RD |
242 | repost: |
243 | if (unlikely(ipoib_ib_post_receive(dev, wr_id))) | |
244 | ipoib_warn(priv, "ipoib_ib_post_receive failed " | |
245 | "for buf %d\n", wr_id); | |
246 | } | |
1da177e4 | 247 | |
7143740d EC |
248 | static int ipoib_dma_map_tx(struct ib_device *ca, |
249 | struct ipoib_tx_buf *tx_req) | |
250 | { | |
251 | struct sk_buff *skb = tx_req->skb; | |
252 | u64 *mapping = tx_req->mapping; | |
253 | int i; | |
40ca1988 | 254 | int off; |
7143740d | 255 | |
40ca1988 EC |
256 | if (skb_headlen(skb)) { |
257 | mapping[0] = ib_dma_map_single(ca, skb->data, skb_headlen(skb), | |
258 | DMA_TO_DEVICE); | |
259 | if (unlikely(ib_dma_mapping_error(ca, mapping[0]))) | |
260 | return -EIO; | |
261 | ||
262 | off = 1; | |
263 | } else | |
264 | off = 0; | |
7143740d EC |
265 | |
266 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
267 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
40ca1988 | 268 | mapping[i + off] = ib_dma_map_page(ca, frag->page, |
7143740d EC |
269 | frag->page_offset, frag->size, |
270 | DMA_TO_DEVICE); | |
40ca1988 | 271 | if (unlikely(ib_dma_mapping_error(ca, mapping[i + off]))) |
7143740d EC |
272 | goto partial_error; |
273 | } | |
274 | return 0; | |
275 | ||
276 | partial_error: | |
7143740d EC |
277 | for (; i > 0; --i) { |
278 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; | |
40ca1988 | 279 | ib_dma_unmap_page(ca, mapping[i - !off], frag->size, DMA_TO_DEVICE); |
7143740d | 280 | } |
40ca1988 EC |
281 | |
282 | if (off) | |
283 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
284 | ||
7143740d EC |
285 | return -EIO; |
286 | } | |
287 | ||
288 | static void ipoib_dma_unmap_tx(struct ib_device *ca, | |
289 | struct ipoib_tx_buf *tx_req) | |
290 | { | |
291 | struct sk_buff *skb = tx_req->skb; | |
292 | u64 *mapping = tx_req->mapping; | |
293 | int i; | |
40ca1988 | 294 | int off; |
7143740d | 295 | |
40ca1988 EC |
296 | if (skb_headlen(skb)) { |
297 | ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE); | |
298 | off = 1; | |
299 | } else | |
300 | off = 0; | |
7143740d EC |
301 | |
302 | for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) { | |
303 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
40ca1988 | 304 | ib_dma_unmap_page(ca, mapping[i + off], frag->size, |
7143740d EC |
305 | DMA_TO_DEVICE); |
306 | } | |
307 | } | |
308 | ||
2439a6e6 RD |
309 | static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc) |
310 | { | |
311 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
312 | unsigned int wr_id = wc->wr_id; | |
313 | struct ipoib_tx_buf *tx_req; | |
314 | unsigned long flags; | |
1da177e4 | 315 | |
a89875fc RD |
316 | ipoib_dbg_data(priv, "send completion: id %d, status: %d\n", |
317 | wr_id, wc->status); | |
1da177e4 | 318 | |
2439a6e6 RD |
319 | if (unlikely(wr_id >= ipoib_sendq_size)) { |
320 | ipoib_warn(priv, "send completion event with wrid %d (> %d)\n", | |
321 | wr_id, ipoib_sendq_size); | |
322 | return; | |
1da177e4 | 323 | } |
2439a6e6 RD |
324 | |
325 | tx_req = &priv->tx_ring[wr_id]; | |
326 | ||
7143740d | 327 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
2439a6e6 | 328 | |
de903512 RD |
329 | ++dev->stats.tx_packets; |
330 | dev->stats.tx_bytes += tx_req->skb->len; | |
2439a6e6 RD |
331 | |
332 | dev_kfree_skb_any(tx_req->skb); | |
333 | ||
334 | spin_lock_irqsave(&priv->tx_lock, flags); | |
335 | ++priv->tx_tail; | |
1b524963 MT |
336 | if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) && |
337 | netif_queue_stopped(dev) && | |
338 | test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) | |
2439a6e6 RD |
339 | netif_wake_queue(dev); |
340 | spin_unlock_irqrestore(&priv->tx_lock, flags); | |
341 | ||
342 | if (wc->status != IB_WC_SUCCESS && | |
343 | wc->status != IB_WC_WR_FLUSH_ERR) | |
344 | ipoib_warn(priv, "failed send event " | |
345 | "(status=%d, wrid=%d vend_err %x)\n", | |
346 | wc->status, wr_id, wc->vendor_err); | |
347 | } | |
348 | ||
bea3348e | 349 | int ipoib_poll(struct napi_struct *napi, int budget) |
2439a6e6 | 350 | { |
bea3348e SH |
351 | struct ipoib_dev_priv *priv = container_of(napi, struct ipoib_dev_priv, napi); |
352 | struct net_device *dev = priv->dev; | |
8d1cc86a RD |
353 | int done; |
354 | int t; | |
8d1cc86a RD |
355 | int n, i; |
356 | ||
357 | done = 0; | |
8d1cc86a | 358 | |
bea3348e SH |
359 | poll_more: |
360 | while (done < budget) { | |
361 | int max = (budget - done); | |
362 | ||
8d1cc86a RD |
363 | t = min(IPOIB_NUM_WC, max); |
364 | n = ib_poll_cq(priv->cq, t, priv->ibwc); | |
365 | ||
bea3348e | 366 | for (i = 0; i < n; i++) { |
8d1cc86a RD |
367 | struct ib_wc *wc = priv->ibwc + i; |
368 | ||
1b524963 | 369 | if (wc->wr_id & IPOIB_OP_RECV) { |
8d1cc86a | 370 | ++done; |
1b524963 MT |
371 | if (wc->wr_id & IPOIB_OP_CM) |
372 | ipoib_cm_handle_rx_wc(dev, wc); | |
373 | else | |
374 | ipoib_ib_handle_rx_wc(dev, wc); | |
375 | } else { | |
376 | if (wc->wr_id & IPOIB_OP_CM) | |
377 | ipoib_cm_handle_tx_wc(dev, wc); | |
378 | else | |
379 | ipoib_ib_handle_tx_wc(dev, wc); | |
380 | } | |
8d1cc86a RD |
381 | } |
382 | ||
bea3348e | 383 | if (n != t) |
8d1cc86a | 384 | break; |
8d1cc86a RD |
385 | } |
386 | ||
bea3348e SH |
387 | if (done < budget) { |
388 | netif_rx_complete(dev, napi); | |
8d1cc86a RD |
389 | if (unlikely(ib_req_notify_cq(priv->cq, |
390 | IB_CQ_NEXT_COMP | | |
391 | IB_CQ_REPORT_MISSED_EVENTS)) && | |
bea3348e SH |
392 | netif_rx_reschedule(dev, napi)) |
393 | goto poll_more; | |
8d1cc86a RD |
394 | } |
395 | ||
bea3348e | 396 | return done; |
1da177e4 LT |
397 | } |
398 | ||
399 | void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr) | |
400 | { | |
bea3348e SH |
401 | struct net_device *dev = dev_ptr; |
402 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
403 | ||
404 | netif_rx_schedule(dev, &priv->napi); | |
1da177e4 LT |
405 | } |
406 | ||
407 | static inline int post_send(struct ipoib_dev_priv *priv, | |
408 | unsigned int wr_id, | |
409 | struct ib_ah *address, u32 qpn, | |
40ca1988 EC |
410 | struct ipoib_tx_buf *tx_req, |
411 | void *head, int hlen) | |
1da177e4 LT |
412 | { |
413 | struct ib_send_wr *bad_wr; | |
40ca1988 EC |
414 | int i, off; |
415 | struct sk_buff *skb = tx_req->skb; | |
416 | skb_frag_t *frags = skb_shinfo(skb)->frags; | |
417 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
418 | u64 *mapping = tx_req->mapping; | |
419 | ||
420 | if (skb_headlen(skb)) { | |
421 | priv->tx_sge[0].addr = mapping[0]; | |
422 | priv->tx_sge[0].length = skb_headlen(skb); | |
423 | off = 1; | |
424 | } else | |
425 | off = 0; | |
1da177e4 | 426 | |
7143740d | 427 | for (i = 0; i < nr_frags; ++i) { |
40ca1988 EC |
428 | priv->tx_sge[i + off].addr = mapping[i + off]; |
429 | priv->tx_sge[i + off].length = frags[i].size; | |
7143740d | 430 | } |
40ca1988 | 431 | priv->tx_wr.num_sge = nr_frags + off; |
7143740d EC |
432 | priv->tx_wr.wr_id = wr_id; |
433 | priv->tx_wr.wr.ud.remote_qpn = qpn; | |
434 | priv->tx_wr.wr.ud.ah = address; | |
1da177e4 | 435 | |
40ca1988 EC |
436 | if (head) { |
437 | priv->tx_wr.wr.ud.mss = skb_shinfo(skb)->gso_size; | |
438 | priv->tx_wr.wr.ud.header = head; | |
439 | priv->tx_wr.wr.ud.hlen = hlen; | |
440 | priv->tx_wr.opcode = IB_WR_LSO; | |
441 | } else | |
442 | priv->tx_wr.opcode = IB_WR_SEND; | |
443 | ||
1da177e4 LT |
444 | return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr); |
445 | } | |
446 | ||
447 | void ipoib_send(struct net_device *dev, struct sk_buff *skb, | |
448 | struct ipoib_ah *address, u32 qpn) | |
449 | { | |
450 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1993d683 | 451 | struct ipoib_tx_buf *tx_req; |
40ca1988 EC |
452 | int hlen; |
453 | void *phead; | |
454 | ||
455 | if (skb_is_gso(skb)) { | |
456 | hlen = skb_transport_offset(skb) + tcp_hdrlen(skb); | |
457 | phead = skb->data; | |
458 | if (unlikely(!skb_pull(skb, hlen))) { | |
459 | ipoib_warn(priv, "linear data too small\n"); | |
460 | ++dev->stats.tx_dropped; | |
461 | ++dev->stats.tx_errors; | |
462 | dev_kfree_skb_any(skb); | |
463 | return; | |
464 | } | |
465 | } else { | |
466 | if (unlikely(skb->len > priv->mcast_mtu + IPOIB_ENCAP_LEN)) { | |
467 | ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n", | |
468 | skb->len, priv->mcast_mtu + IPOIB_ENCAP_LEN); | |
469 | ++dev->stats.tx_dropped; | |
470 | ++dev->stats.tx_errors; | |
471 | ipoib_cm_skb_too_long(dev, skb, priv->mcast_mtu); | |
472 | return; | |
473 | } | |
474 | phead = NULL; | |
475 | hlen = 0; | |
1da177e4 LT |
476 | } |
477 | ||
478 | ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n", | |
479 | skb->len, address, qpn); | |
480 | ||
481 | /* | |
482 | * We put the skb into the tx_ring _before_ we call post_send() | |
483 | * because it's entirely possible that the completion handler will | |
484 | * run before we execute anything after the post_send(). That | |
485 | * means we have to make sure everything is properly recorded and | |
486 | * our state is consistent before we call post_send(). | |
487 | */ | |
0f485251 | 488 | tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)]; |
1da177e4 | 489 | tx_req->skb = skb; |
7143740d | 490 | if (unlikely(ipoib_dma_map_tx(priv->ca, tx_req))) { |
de903512 | 491 | ++dev->stats.tx_errors; |
73fbe8be RD |
492 | dev_kfree_skb_any(skb); |
493 | return; | |
494 | } | |
1da177e4 | 495 | |
6046136c EC |
496 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
497 | priv->tx_wr.send_flags |= IB_SEND_IP_CSUM; | |
498 | else | |
499 | priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM; | |
500 | ||
0f485251 | 501 | if (unlikely(post_send(priv, priv->tx_head & (ipoib_sendq_size - 1), |
40ca1988 | 502 | address->ah, qpn, tx_req, phead, hlen))) { |
1da177e4 | 503 | ipoib_warn(priv, "post_send failed\n"); |
de903512 | 504 | ++dev->stats.tx_errors; |
7143740d | 505 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
1da177e4 LT |
506 | dev_kfree_skb_any(skb); |
507 | } else { | |
508 | dev->trans_start = jiffies; | |
509 | ||
510 | address->last_send = priv->tx_head; | |
511 | ++priv->tx_head; | |
512 | ||
1b524963 | 513 | if (++priv->tx_outstanding == ipoib_sendq_size) { |
1da177e4 LT |
514 | ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n"); |
515 | netif_stop_queue(dev); | |
516 | } | |
517 | } | |
518 | } | |
519 | ||
520 | static void __ipoib_reap_ah(struct net_device *dev) | |
521 | { | |
522 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
523 | struct ipoib_ah *ah, *tah; | |
524 | LIST_HEAD(remove_list); | |
525 | ||
31c02e21 RD |
526 | spin_lock_irq(&priv->tx_lock); |
527 | spin_lock(&priv->lock); | |
1da177e4 | 528 | list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list) |
2181858b | 529 | if ((int) priv->tx_tail - (int) ah->last_send >= 0) { |
1da177e4 | 530 | list_del(&ah->list); |
31c02e21 RD |
531 | ib_destroy_ah(ah->ah); |
532 | kfree(ah); | |
1da177e4 | 533 | } |
31c02e21 RD |
534 | spin_unlock(&priv->lock); |
535 | spin_unlock_irq(&priv->tx_lock); | |
1da177e4 LT |
536 | } |
537 | ||
c4028958 | 538 | void ipoib_reap_ah(struct work_struct *work) |
1da177e4 | 539 | { |
c4028958 DH |
540 | struct ipoib_dev_priv *priv = |
541 | container_of(work, struct ipoib_dev_priv, ah_reap_task.work); | |
542 | struct net_device *dev = priv->dev; | |
1da177e4 LT |
543 | |
544 | __ipoib_reap_ah(dev); | |
545 | ||
546 | if (!test_bit(IPOIB_STOP_REAPER, &priv->flags)) | |
69fc507a AB |
547 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, |
548 | round_jiffies_relative(HZ)); | |
1da177e4 LT |
549 | } |
550 | ||
551 | int ipoib_ib_dev_open(struct net_device *dev) | |
552 | { | |
553 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
554 | int ret; | |
555 | ||
26bbf13c YE |
556 | if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &priv->pkey_index)) { |
557 | ipoib_warn(priv, "P_Key 0x%04x not found\n", priv->pkey); | |
558 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
559 | return -1; | |
560 | } | |
561 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
562 | ||
5b6810e0 | 563 | ret = ipoib_init_qp(dev); |
1da177e4 | 564 | if (ret) { |
5b6810e0 | 565 | ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret); |
1da177e4 LT |
566 | return -1; |
567 | } | |
568 | ||
569 | ret = ipoib_ib_post_receives(dev); | |
570 | if (ret) { | |
571 | ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret); | |
26bbf13c | 572 | ipoib_ib_dev_stop(dev, 1); |
1da177e4 LT |
573 | return -1; |
574 | } | |
575 | ||
839fcaba MT |
576 | ret = ipoib_cm_dev_open(dev); |
577 | if (ret) { | |
24bd1e4e | 578 | ipoib_warn(priv, "ipoib_cm_dev_open returned %d\n", ret); |
26bbf13c | 579 | ipoib_ib_dev_stop(dev, 1); |
839fcaba MT |
580 | return -1; |
581 | } | |
582 | ||
1da177e4 | 583 | clear_bit(IPOIB_STOP_REAPER, &priv->flags); |
69fc507a AB |
584 | queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, |
585 | round_jiffies_relative(HZ)); | |
1da177e4 | 586 | |
7a343d4c LA |
587 | set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags); |
588 | ||
1da177e4 LT |
589 | return 0; |
590 | } | |
591 | ||
7a343d4c LA |
592 | static void ipoib_pkey_dev_check_presence(struct net_device *dev) |
593 | { | |
594 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
595 | u16 pkey_index = 0; | |
596 | ||
597 | if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) | |
598 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
599 | else | |
600 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
601 | } | |
602 | ||
1da177e4 LT |
603 | int ipoib_ib_dev_up(struct net_device *dev) |
604 | { | |
605 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
606 | ||
7a343d4c LA |
607 | ipoib_pkey_dev_check_presence(dev); |
608 | ||
609 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
610 | ipoib_dbg(priv, "PKEY is not assigned.\n"); | |
611 | return 0; | |
612 | } | |
613 | ||
1da177e4 LT |
614 | set_bit(IPOIB_FLAG_OPER_UP, &priv->flags); |
615 | ||
616 | return ipoib_mcast_start_thread(dev); | |
617 | } | |
618 | ||
0b3ea082 | 619 | int ipoib_ib_dev_down(struct net_device *dev, int flush) |
1da177e4 LT |
620 | { |
621 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
622 | ||
623 | ipoib_dbg(priv, "downing ib_dev\n"); | |
624 | ||
625 | clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags); | |
626 | netif_carrier_off(dev); | |
627 | ||
628 | /* Shutdown the P_Key thread if still active */ | |
629 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
95ed644f | 630 | mutex_lock(&pkey_mutex); |
1da177e4 | 631 | set_bit(IPOIB_PKEY_STOP, &priv->flags); |
26bbf13c | 632 | cancel_delayed_work(&priv->pkey_poll_task); |
95ed644f | 633 | mutex_unlock(&pkey_mutex); |
0b3ea082 JM |
634 | if (flush) |
635 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
636 | } |
637 | ||
0b3ea082 | 638 | ipoib_mcast_stop_thread(dev, flush); |
1da177e4 LT |
639 | ipoib_mcast_dev_flush(dev); |
640 | ||
1da177e4 LT |
641 | ipoib_flush_paths(dev); |
642 | ||
643 | return 0; | |
644 | } | |
645 | ||
646 | static int recvs_pending(struct net_device *dev) | |
647 | { | |
648 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
649 | int pending = 0; | |
650 | int i; | |
651 | ||
0f485251 | 652 | for (i = 0; i < ipoib_recvq_size; ++i) |
1da177e4 LT |
653 | if (priv->rx_ring[i].skb) |
654 | ++pending; | |
655 | ||
656 | return pending; | |
657 | } | |
658 | ||
2dfbfc37 MT |
659 | void ipoib_drain_cq(struct net_device *dev) |
660 | { | |
661 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
662 | int i, n; | |
663 | do { | |
664 | n = ib_poll_cq(priv->cq, IPOIB_NUM_WC, priv->ibwc); | |
665 | for (i = 0; i < n; ++i) { | |
ce423ef5 RD |
666 | /* |
667 | * Convert any successful completions to flush | |
668 | * errors to avoid passing packets up the | |
669 | * stack after bringing the device down. | |
670 | */ | |
671 | if (priv->ibwc[i].status == IB_WC_SUCCESS) | |
672 | priv->ibwc[i].status = IB_WC_WR_FLUSH_ERR; | |
673 | ||
1b524963 MT |
674 | if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) { |
675 | if (priv->ibwc[i].wr_id & IPOIB_OP_CM) | |
676 | ipoib_cm_handle_rx_wc(dev, priv->ibwc + i); | |
677 | else | |
678 | ipoib_ib_handle_rx_wc(dev, priv->ibwc + i); | |
679 | } else { | |
680 | if (priv->ibwc[i].wr_id & IPOIB_OP_CM) | |
681 | ipoib_cm_handle_tx_wc(dev, priv->ibwc + i); | |
682 | else | |
683 | ipoib_ib_handle_tx_wc(dev, priv->ibwc + i); | |
684 | } | |
2dfbfc37 MT |
685 | } |
686 | } while (n == IPOIB_NUM_WC); | |
687 | } | |
688 | ||
26bbf13c | 689 | int ipoib_ib_dev_stop(struct net_device *dev, int flush) |
1da177e4 LT |
690 | { |
691 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
692 | struct ib_qp_attr qp_attr; | |
1da177e4 | 693 | unsigned long begin; |
1993d683 | 694 | struct ipoib_tx_buf *tx_req; |
2dfbfc37 | 695 | int i; |
1da177e4 | 696 | |
7a343d4c LA |
697 | clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags); |
698 | ||
839fcaba MT |
699 | ipoib_cm_dev_stop(dev); |
700 | ||
3bc12e75 RD |
701 | /* |
702 | * Move our QP to the error state and then reinitialize in | |
703 | * when all work requests have completed or have been flushed. | |
704 | */ | |
1da177e4 | 705 | qp_attr.qp_state = IB_QPS_ERR; |
3bc12e75 | 706 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
707 | ipoib_warn(priv, "Failed to modify QP to ERROR state\n"); |
708 | ||
709 | /* Wait for all sends and receives to complete */ | |
710 | begin = jiffies; | |
711 | ||
712 | while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) { | |
713 | if (time_after(jiffies, begin + 5 * HZ)) { | |
714 | ipoib_warn(priv, "timing out; %d sends %d receives not completed\n", | |
715 | priv->tx_head - priv->tx_tail, recvs_pending(dev)); | |
716 | ||
717 | /* | |
718 | * assume the HW is wedged and just free up | |
719 | * all our pending work requests. | |
720 | */ | |
2181858b | 721 | while ((int) priv->tx_tail - (int) priv->tx_head < 0) { |
1da177e4 | 722 | tx_req = &priv->tx_ring[priv->tx_tail & |
0f485251 | 723 | (ipoib_sendq_size - 1)]; |
7143740d | 724 | ipoib_dma_unmap_tx(priv->ca, tx_req); |
1da177e4 LT |
725 | dev_kfree_skb_any(tx_req->skb); |
726 | ++priv->tx_tail; | |
1b524963 | 727 | --priv->tx_outstanding; |
1da177e4 LT |
728 | } |
729 | ||
37ccf9df RC |
730 | for (i = 0; i < ipoib_recvq_size; ++i) { |
731 | struct ipoib_rx_buf *rx_req; | |
732 | ||
733 | rx_req = &priv->rx_ring[i]; | |
734 | if (!rx_req->skb) | |
735 | continue; | |
736 | ib_dma_unmap_single(priv->ca, | |
737 | rx_req->mapping, | |
738 | IPOIB_BUF_SIZE, | |
739 | DMA_FROM_DEVICE); | |
740 | dev_kfree_skb_any(rx_req->skb); | |
741 | rx_req->skb = NULL; | |
742 | } | |
1da177e4 LT |
743 | |
744 | goto timeout; | |
745 | } | |
746 | ||
2dfbfc37 | 747 | ipoib_drain_cq(dev); |
8d1cc86a | 748 | |
1da177e4 LT |
749 | msleep(1); |
750 | } | |
751 | ||
752 | ipoib_dbg(priv, "All sends and receives done.\n"); | |
753 | ||
754 | timeout: | |
755 | qp_attr.qp_state = IB_QPS_RESET; | |
3bc12e75 | 756 | if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE)) |
1da177e4 LT |
757 | ipoib_warn(priv, "Failed to modify QP to RESET state\n"); |
758 | ||
759 | /* Wait for all AHs to be reaped */ | |
760 | set_bit(IPOIB_STOP_REAPER, &priv->flags); | |
761 | cancel_delayed_work(&priv->ah_reap_task); | |
26bbf13c YE |
762 | if (flush) |
763 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
764 | |
765 | begin = jiffies; | |
766 | ||
767 | while (!list_empty(&priv->dead_ahs)) { | |
768 | __ipoib_reap_ah(dev); | |
769 | ||
770 | if (time_after(jiffies, begin + HZ)) { | |
771 | ipoib_warn(priv, "timing out; will leak address handles\n"); | |
772 | break; | |
773 | } | |
774 | ||
775 | msleep(1); | |
776 | } | |
777 | ||
8d1cc86a RD |
778 | ib_req_notify_cq(priv->cq, IB_CQ_NEXT_COMP); |
779 | ||
1da177e4 LT |
780 | return 0; |
781 | } | |
782 | ||
783 | int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port) | |
784 | { | |
785 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
786 | ||
787 | priv->ca = ca; | |
788 | priv->port = port; | |
789 | priv->qp = NULL; | |
790 | ||
791 | if (ipoib_transport_dev_init(dev, ca)) { | |
792 | printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name); | |
793 | return -ENODEV; | |
794 | } | |
795 | ||
796 | if (dev->flags & IFF_UP) { | |
797 | if (ipoib_ib_dev_open(dev)) { | |
798 | ipoib_transport_dev_cleanup(dev); | |
799 | return -ENODEV; | |
800 | } | |
801 | } | |
802 | ||
803 | return 0; | |
804 | } | |
805 | ||
26bbf13c | 806 | static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv, int pkey_event) |
1da177e4 | 807 | { |
26bbf13c | 808 | struct ipoib_dev_priv *cpriv; |
c4028958 | 809 | struct net_device *dev = priv->dev; |
26bbf13c YE |
810 | u16 new_index; |
811 | ||
812 | mutex_lock(&priv->vlan_mutex); | |
1da177e4 | 813 | |
26bbf13c YE |
814 | /* |
815 | * Flush any child interfaces too -- they might be up even if | |
816 | * the parent is down. | |
817 | */ | |
818 | list_for_each_entry(cpriv, &priv->child_intfs, list) | |
819 | __ipoib_ib_dev_flush(cpriv, pkey_event); | |
820 | ||
821 | mutex_unlock(&priv->vlan_mutex); | |
822 | ||
823 | if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) { | |
7a343d4c LA |
824 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n"); |
825 | return; | |
826 | } | |
827 | ||
828 | if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { | |
829 | ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n"); | |
1da177e4 | 830 | return; |
7a343d4c | 831 | } |
1da177e4 | 832 | |
26bbf13c YE |
833 | if (pkey_event) { |
834 | if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &new_index)) { | |
835 | clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
836 | ipoib_ib_dev_down(dev, 0); | |
167c4265 | 837 | ipoib_ib_dev_stop(dev, 0); |
26bbf13c YE |
838 | ipoib_pkey_dev_delay_open(dev); |
839 | return; | |
840 | } | |
841 | set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags); | |
842 | ||
843 | /* restart QP only if P_Key index is changed */ | |
844 | if (new_index == priv->pkey_index) { | |
845 | ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n"); | |
846 | return; | |
847 | } | |
848 | priv->pkey_index = new_index; | |
849 | } | |
850 | ||
1da177e4 LT |
851 | ipoib_dbg(priv, "flushing\n"); |
852 | ||
0b3ea082 | 853 | ipoib_ib_dev_down(dev, 0); |
1da177e4 | 854 | |
26bbf13c YE |
855 | if (pkey_event) { |
856 | ipoib_ib_dev_stop(dev, 0); | |
857 | ipoib_ib_dev_open(dev); | |
858 | } | |
859 | ||
1da177e4 LT |
860 | /* |
861 | * The device could have been brought down between the start and when | |
862 | * we get here, don't bring it back up if it's not configured up | |
863 | */ | |
5ccd0255 | 864 | if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) { |
1da177e4 | 865 | ipoib_ib_dev_up(dev); |
c4028958 | 866 | ipoib_mcast_restart_task(&priv->restart_task); |
5ccd0255 | 867 | } |
26bbf13c | 868 | } |
1da177e4 | 869 | |
26bbf13c YE |
870 | void ipoib_ib_dev_flush(struct work_struct *work) |
871 | { | |
872 | struct ipoib_dev_priv *priv = | |
873 | container_of(work, struct ipoib_dev_priv, flush_task); | |
4f71055a | 874 | |
26bbf13c YE |
875 | ipoib_dbg(priv, "Flushing %s\n", priv->dev->name); |
876 | __ipoib_ib_dev_flush(priv, 0); | |
877 | } | |
4f71055a | 878 | |
26bbf13c YE |
879 | void ipoib_pkey_event(struct work_struct *work) |
880 | { | |
881 | struct ipoib_dev_priv *priv = | |
882 | container_of(work, struct ipoib_dev_priv, pkey_event_task); | |
883 | ||
884 | ipoib_dbg(priv, "Flushing %s and restarting its QP\n", priv->dev->name); | |
885 | __ipoib_ib_dev_flush(priv, 1); | |
1da177e4 LT |
886 | } |
887 | ||
888 | void ipoib_ib_dev_cleanup(struct net_device *dev) | |
889 | { | |
890 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
891 | ||
892 | ipoib_dbg(priv, "cleaning up ib_dev\n"); | |
893 | ||
8d2cae06 | 894 | ipoib_mcast_stop_thread(dev, 1); |
988bd503 | 895 | ipoib_mcast_dev_flush(dev); |
1da177e4 LT |
896 | |
897 | ipoib_transport_dev_cleanup(dev); | |
898 | } | |
899 | ||
900 | /* | |
901 | * Delayed P_Key Assigment Interim Support | |
902 | * | |
903 | * The following is initial implementation of delayed P_Key assigment | |
904 | * mechanism. It is using the same approach implemented for the multicast | |
905 | * group join. The single goal of this implementation is to quickly address | |
906 | * Bug #2507. This implementation will probably be removed when the P_Key | |
907 | * change async notification is available. | |
908 | */ | |
1da177e4 | 909 | |
c4028958 | 910 | void ipoib_pkey_poll(struct work_struct *work) |
1da177e4 | 911 | { |
c4028958 | 912 | struct ipoib_dev_priv *priv = |
26bbf13c | 913 | container_of(work, struct ipoib_dev_priv, pkey_poll_task.work); |
c4028958 | 914 | struct net_device *dev = priv->dev; |
1da177e4 LT |
915 | |
916 | ipoib_pkey_dev_check_presence(dev); | |
917 | ||
918 | if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) | |
919 | ipoib_open(dev); | |
920 | else { | |
95ed644f | 921 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
922 | if (!test_bit(IPOIB_PKEY_STOP, &priv->flags)) |
923 | queue_delayed_work(ipoib_workqueue, | |
26bbf13c | 924 | &priv->pkey_poll_task, |
1da177e4 | 925 | HZ); |
95ed644f | 926 | mutex_unlock(&pkey_mutex); |
1da177e4 LT |
927 | } |
928 | } | |
929 | ||
930 | int ipoib_pkey_dev_delay_open(struct net_device *dev) | |
931 | { | |
932 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
933 | ||
934 | /* Look for the interface pkey value in the IB Port P_Key table and */ | |
935 | /* set the interface pkey assigment flag */ | |
936 | ipoib_pkey_dev_check_presence(dev); | |
937 | ||
938 | /* P_Key value not assigned yet - start polling */ | |
939 | if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) { | |
95ed644f | 940 | mutex_lock(&pkey_mutex); |
1da177e4 LT |
941 | clear_bit(IPOIB_PKEY_STOP, &priv->flags); |
942 | queue_delayed_work(ipoib_workqueue, | |
26bbf13c | 943 | &priv->pkey_poll_task, |
1da177e4 | 944 | HZ); |
95ed644f | 945 | mutex_unlock(&pkey_mutex); |
1da177e4 LT |
946 | return 1; |
947 | } | |
948 | ||
949 | return 0; | |
950 | } |