ipoib_ib.c 18 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709
  1. /*
  2. * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
  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.
  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. #include <linux/delay.h>
  38. #include <linux/dma-mapping.h>
  39. #include <rdma/ib_cache.h>
  40. #include "ipoib.h"
  41. #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
  42. static int data_debug_level;
  43. module_param(data_debug_level, int, 0644);
  44. MODULE_PARM_DESC(data_debug_level,
  45. "Enable data path debug tracing if > 0");
  46. #endif
  47. #define IPOIB_OP_RECV (1ul << 31)
  48. static DEFINE_MUTEX(pkey_mutex);
  49. struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
  50. struct ib_pd *pd, struct ib_ah_attr *attr)
  51. {
  52. struct ipoib_ah *ah;
  53. ah = kmalloc(sizeof *ah, GFP_KERNEL);
  54. if (!ah)
  55. return NULL;
  56. ah->dev = dev;
  57. ah->last_send = 0;
  58. kref_init(&ah->ref);
  59. ah->ah = ib_create_ah(pd, attr);
  60. if (IS_ERR(ah->ah)) {
  61. kfree(ah);
  62. ah = NULL;
  63. } else
  64. ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah);
  65. return ah;
  66. }
  67. void ipoib_free_ah(struct kref *kref)
  68. {
  69. struct ipoib_ah *ah = container_of(kref, struct ipoib_ah, ref);
  70. struct ipoib_dev_priv *priv = netdev_priv(ah->dev);
  71. unsigned long flags;
  72. spin_lock_irqsave(&priv->lock, flags);
  73. list_add_tail(&ah->list, &priv->dead_ahs);
  74. spin_unlock_irqrestore(&priv->lock, flags);
  75. }
  76. static int ipoib_ib_post_receive(struct net_device *dev, int id)
  77. {
  78. struct ipoib_dev_priv *priv = netdev_priv(dev);
  79. struct ib_sge list;
  80. struct ib_recv_wr param;
  81. struct ib_recv_wr *bad_wr;
  82. int ret;
  83. list.addr = priv->rx_ring[id].mapping;
  84. list.length = IPOIB_BUF_SIZE;
  85. list.lkey = priv->mr->lkey;
  86. param.next = NULL;
  87. param.wr_id = id | IPOIB_OP_RECV;
  88. param.sg_list = &list;
  89. param.num_sge = 1;
  90. ret = ib_post_recv(priv->qp, &param, &bad_wr);
  91. if (unlikely(ret)) {
  92. ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret);
  93. dma_unmap_single(priv->ca->dma_device,
  94. priv->rx_ring[id].mapping,
  95. IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
  96. dev_kfree_skb_any(priv->rx_ring[id].skb);
  97. priv->rx_ring[id].skb = NULL;
  98. }
  99. return ret;
  100. }
  101. static int ipoib_alloc_rx_skb(struct net_device *dev, int id)
  102. {
  103. struct ipoib_dev_priv *priv = netdev_priv(dev);
  104. struct sk_buff *skb;
  105. dma_addr_t addr;
  106. skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4);
  107. if (!skb)
  108. return -ENOMEM;
  109. /*
  110. * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte
  111. * header. So we need 4 more bytes to get to 48 and align the
  112. * IP header to a multiple of 16.
  113. */
  114. skb_reserve(skb, 4);
  115. addr = dma_map_single(priv->ca->dma_device,
  116. skb->data, IPOIB_BUF_SIZE,
  117. DMA_FROM_DEVICE);
  118. if (unlikely(dma_mapping_error(addr))) {
  119. dev_kfree_skb_any(skb);
  120. return -EIO;
  121. }
  122. priv->rx_ring[id].skb = skb;
  123. priv->rx_ring[id].mapping = addr;
  124. return 0;
  125. }
  126. static int ipoib_ib_post_receives(struct net_device *dev)
  127. {
  128. struct ipoib_dev_priv *priv = netdev_priv(dev);
  129. int i;
  130. for (i = 0; i < ipoib_recvq_size; ++i) {
  131. if (ipoib_alloc_rx_skb(dev, i)) {
  132. ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
  133. return -ENOMEM;
  134. }
  135. if (ipoib_ib_post_receive(dev, i)) {
  136. ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i);
  137. return -EIO;
  138. }
  139. }
  140. return 0;
  141. }
  142. static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
  143. {
  144. struct ipoib_dev_priv *priv = netdev_priv(dev);
  145. unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV;
  146. struct sk_buff *skb;
  147. dma_addr_t addr;
  148. ipoib_dbg_data(priv, "recv completion: id %d, op %d, status: %d\n",
  149. wr_id, wc->opcode, wc->status);
  150. if (unlikely(wr_id >= ipoib_recvq_size)) {
  151. ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n",
  152. wr_id, ipoib_recvq_size);
  153. return;
  154. }
  155. skb = priv->rx_ring[wr_id].skb;
  156. addr = priv->rx_ring[wr_id].mapping;
  157. if (unlikely(wc->status != IB_WC_SUCCESS)) {
  158. if (wc->status != IB_WC_WR_FLUSH_ERR)
  159. ipoib_warn(priv, "failed recv event "
  160. "(status=%d, wrid=%d vend_err %x)\n",
  161. wc->status, wr_id, wc->vendor_err);
  162. dma_unmap_single(priv->ca->dma_device, addr,
  163. IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
  164. dev_kfree_skb_any(skb);
  165. priv->rx_ring[wr_id].skb = NULL;
  166. return;
  167. }
  168. /*
  169. * If we can't allocate a new RX buffer, dump
  170. * this packet and reuse the old buffer.
  171. */
  172. if (unlikely(ipoib_alloc_rx_skb(dev, wr_id))) {
  173. ++priv->stats.rx_dropped;
  174. goto repost;
  175. }
  176. ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
  177. wc->byte_len, wc->slid);
  178. dma_unmap_single(priv->ca->dma_device, addr,
  179. IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
  180. skb_put(skb, wc->byte_len);
  181. skb_pull(skb, IB_GRH_BYTES);
  182. if (wc->slid != priv->local_lid ||
  183. wc->src_qp != priv->qp->qp_num) {
  184. skb->protocol = ((struct ipoib_header *) skb->data)->proto;
  185. skb->mac.raw = skb->data;
  186. skb_pull(skb, IPOIB_ENCAP_LEN);
  187. dev->last_rx = jiffies;
  188. ++priv->stats.rx_packets;
  189. priv->stats.rx_bytes += skb->len;
  190. skb->dev = dev;
  191. /* XXX get correct PACKET_ type here */
  192. skb->pkt_type = PACKET_HOST;
  193. netif_rx_ni(skb);
  194. } else {
  195. ipoib_dbg_data(priv, "dropping loopback packet\n");
  196. dev_kfree_skb_any(skb);
  197. }
  198. repost:
  199. if (unlikely(ipoib_ib_post_receive(dev, wr_id)))
  200. ipoib_warn(priv, "ipoib_ib_post_receive failed "
  201. "for buf %d\n", wr_id);
  202. }
  203. static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
  204. {
  205. struct ipoib_dev_priv *priv = netdev_priv(dev);
  206. unsigned int wr_id = wc->wr_id;
  207. struct ipoib_tx_buf *tx_req;
  208. unsigned long flags;
  209. ipoib_dbg_data(priv, "send completion: id %d, op %d, status: %d\n",
  210. wr_id, wc->opcode, wc->status);
  211. if (unlikely(wr_id >= ipoib_sendq_size)) {
  212. ipoib_warn(priv, "send completion event with wrid %d (> %d)\n",
  213. wr_id, ipoib_sendq_size);
  214. return;
  215. }
  216. tx_req = &priv->tx_ring[wr_id];
  217. dma_unmap_single(priv->ca->dma_device,
  218. pci_unmap_addr(tx_req, mapping),
  219. tx_req->skb->len,
  220. DMA_TO_DEVICE);
  221. ++priv->stats.tx_packets;
  222. priv->stats.tx_bytes += tx_req->skb->len;
  223. dev_kfree_skb_any(tx_req->skb);
  224. spin_lock_irqsave(&priv->tx_lock, flags);
  225. ++priv->tx_tail;
  226. if (netif_queue_stopped(dev) &&
  227. test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags) &&
  228. priv->tx_head - priv->tx_tail <= ipoib_sendq_size >> 1)
  229. netif_wake_queue(dev);
  230. spin_unlock_irqrestore(&priv->tx_lock, flags);
  231. if (wc->status != IB_WC_SUCCESS &&
  232. wc->status != IB_WC_WR_FLUSH_ERR)
  233. ipoib_warn(priv, "failed send event "
  234. "(status=%d, wrid=%d vend_err %x)\n",
  235. wc->status, wr_id, wc->vendor_err);
  236. }
  237. static void ipoib_ib_handle_wc(struct net_device *dev, struct ib_wc *wc)
  238. {
  239. if (wc->wr_id & IPOIB_OP_RECV)
  240. ipoib_ib_handle_rx_wc(dev, wc);
  241. else
  242. ipoib_ib_handle_tx_wc(dev, wc);
  243. }
  244. void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
  245. {
  246. struct net_device *dev = (struct net_device *) dev_ptr;
  247. struct ipoib_dev_priv *priv = netdev_priv(dev);
  248. int n, i;
  249. ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
  250. do {
  251. n = ib_poll_cq(cq, IPOIB_NUM_WC, priv->ibwc);
  252. for (i = 0; i < n; ++i)
  253. ipoib_ib_handle_wc(dev, priv->ibwc + i);
  254. } while (n == IPOIB_NUM_WC);
  255. }
  256. static inline int post_send(struct ipoib_dev_priv *priv,
  257. unsigned int wr_id,
  258. struct ib_ah *address, u32 qpn,
  259. dma_addr_t addr, int len)
  260. {
  261. struct ib_send_wr *bad_wr;
  262. priv->tx_sge.addr = addr;
  263. priv->tx_sge.length = len;
  264. priv->tx_wr.wr_id = wr_id;
  265. priv->tx_wr.wr.ud.remote_qpn = qpn;
  266. priv->tx_wr.wr.ud.ah = address;
  267. return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr);
  268. }
  269. void ipoib_send(struct net_device *dev, struct sk_buff *skb,
  270. struct ipoib_ah *address, u32 qpn)
  271. {
  272. struct ipoib_dev_priv *priv = netdev_priv(dev);
  273. struct ipoib_tx_buf *tx_req;
  274. dma_addr_t addr;
  275. if (unlikely(skb->len > dev->mtu + INFINIBAND_ALEN)) {
  276. ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
  277. skb->len, dev->mtu + INFINIBAND_ALEN);
  278. ++priv->stats.tx_dropped;
  279. ++priv->stats.tx_errors;
  280. dev_kfree_skb_any(skb);
  281. return;
  282. }
  283. ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n",
  284. skb->len, address, qpn);
  285. /*
  286. * We put the skb into the tx_ring _before_ we call post_send()
  287. * because it's entirely possible that the completion handler will
  288. * run before we execute anything after the post_send(). That
  289. * means we have to make sure everything is properly recorded and
  290. * our state is consistent before we call post_send().
  291. */
  292. tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)];
  293. tx_req->skb = skb;
  294. addr = dma_map_single(priv->ca->dma_device, skb->data, skb->len,
  295. DMA_TO_DEVICE);
  296. pci_unmap_addr_set(tx_req, mapping, addr);
  297. if (unlikely(post_send(priv, priv->tx_head & (ipoib_sendq_size - 1),
  298. address->ah, qpn, addr, skb->len))) {
  299. ipoib_warn(priv, "post_send failed\n");
  300. ++priv->stats.tx_errors;
  301. dma_unmap_single(priv->ca->dma_device, addr, skb->len,
  302. DMA_TO_DEVICE);
  303. dev_kfree_skb_any(skb);
  304. } else {
  305. dev->trans_start = jiffies;
  306. address->last_send = priv->tx_head;
  307. ++priv->tx_head;
  308. if (priv->tx_head - priv->tx_tail == ipoib_sendq_size) {
  309. ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
  310. netif_stop_queue(dev);
  311. }
  312. }
  313. }
  314. static void __ipoib_reap_ah(struct net_device *dev)
  315. {
  316. struct ipoib_dev_priv *priv = netdev_priv(dev);
  317. struct ipoib_ah *ah, *tah;
  318. LIST_HEAD(remove_list);
  319. spin_lock_irq(&priv->tx_lock);
  320. spin_lock(&priv->lock);
  321. list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list)
  322. if ((int) priv->tx_tail - (int) ah->last_send >= 0) {
  323. list_del(&ah->list);
  324. ib_destroy_ah(ah->ah);
  325. kfree(ah);
  326. }
  327. spin_unlock(&priv->lock);
  328. spin_unlock_irq(&priv->tx_lock);
  329. }
  330. void ipoib_reap_ah(void *dev_ptr)
  331. {
  332. struct net_device *dev = dev_ptr;
  333. struct ipoib_dev_priv *priv = netdev_priv(dev);
  334. __ipoib_reap_ah(dev);
  335. if (!test_bit(IPOIB_STOP_REAPER, &priv->flags))
  336. queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
  337. }
  338. int ipoib_ib_dev_open(struct net_device *dev)
  339. {
  340. struct ipoib_dev_priv *priv = netdev_priv(dev);
  341. int ret;
  342. ret = ipoib_init_qp(dev);
  343. if (ret) {
  344. ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret);
  345. return -1;
  346. }
  347. ret = ipoib_ib_post_receives(dev);
  348. if (ret) {
  349. ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret);
  350. ipoib_ib_dev_stop(dev);
  351. return -1;
  352. }
  353. clear_bit(IPOIB_STOP_REAPER, &priv->flags);
  354. queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
  355. set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags);
  356. return 0;
  357. }
  358. static void ipoib_pkey_dev_check_presence(struct net_device *dev)
  359. {
  360. struct ipoib_dev_priv *priv = netdev_priv(dev);
  361. u16 pkey_index = 0;
  362. if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index))
  363. clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
  364. else
  365. set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
  366. }
  367. int ipoib_ib_dev_up(struct net_device *dev)
  368. {
  369. struct ipoib_dev_priv *priv = netdev_priv(dev);
  370. ipoib_pkey_dev_check_presence(dev);
  371. if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
  372. ipoib_dbg(priv, "PKEY is not assigned.\n");
  373. return 0;
  374. }
  375. set_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
  376. return ipoib_mcast_start_thread(dev);
  377. }
  378. int ipoib_ib_dev_down(struct net_device *dev, int flush)
  379. {
  380. struct ipoib_dev_priv *priv = netdev_priv(dev);
  381. ipoib_dbg(priv, "downing ib_dev\n");
  382. clear_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
  383. netif_carrier_off(dev);
  384. /* Shutdown the P_Key thread if still active */
  385. if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
  386. mutex_lock(&pkey_mutex);
  387. set_bit(IPOIB_PKEY_STOP, &priv->flags);
  388. cancel_delayed_work(&priv->pkey_task);
  389. mutex_unlock(&pkey_mutex);
  390. if (flush)
  391. flush_workqueue(ipoib_workqueue);
  392. }
  393. ipoib_mcast_stop_thread(dev, flush);
  394. ipoib_mcast_dev_flush(dev);
  395. ipoib_flush_paths(dev);
  396. return 0;
  397. }
  398. static int recvs_pending(struct net_device *dev)
  399. {
  400. struct ipoib_dev_priv *priv = netdev_priv(dev);
  401. int pending = 0;
  402. int i;
  403. for (i = 0; i < ipoib_recvq_size; ++i)
  404. if (priv->rx_ring[i].skb)
  405. ++pending;
  406. return pending;
  407. }
  408. int ipoib_ib_dev_stop(struct net_device *dev)
  409. {
  410. struct ipoib_dev_priv *priv = netdev_priv(dev);
  411. struct ib_qp_attr qp_attr;
  412. unsigned long begin;
  413. struct ipoib_tx_buf *tx_req;
  414. int i;
  415. clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags);
  416. /*
  417. * Move our QP to the error state and then reinitialize in
  418. * when all work requests have completed or have been flushed.
  419. */
  420. qp_attr.qp_state = IB_QPS_ERR;
  421. if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
  422. ipoib_warn(priv, "Failed to modify QP to ERROR state\n");
  423. /* Wait for all sends and receives to complete */
  424. begin = jiffies;
  425. while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) {
  426. if (time_after(jiffies, begin + 5 * HZ)) {
  427. ipoib_warn(priv, "timing out; %d sends %d receives not completed\n",
  428. priv->tx_head - priv->tx_tail, recvs_pending(dev));
  429. /*
  430. * assume the HW is wedged and just free up
  431. * all our pending work requests.
  432. */
  433. while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
  434. tx_req = &priv->tx_ring[priv->tx_tail &
  435. (ipoib_sendq_size - 1)];
  436. dma_unmap_single(priv->ca->dma_device,
  437. pci_unmap_addr(tx_req, mapping),
  438. tx_req->skb->len,
  439. DMA_TO_DEVICE);
  440. dev_kfree_skb_any(tx_req->skb);
  441. ++priv->tx_tail;
  442. }
  443. for (i = 0; i < ipoib_recvq_size; ++i)
  444. if (priv->rx_ring[i].skb) {
  445. dma_unmap_single(priv->ca->dma_device,
  446. pci_unmap_addr(&priv->rx_ring[i],
  447. mapping),
  448. IPOIB_BUF_SIZE,
  449. DMA_FROM_DEVICE);
  450. dev_kfree_skb_any(priv->rx_ring[i].skb);
  451. priv->rx_ring[i].skb = NULL;
  452. }
  453. goto timeout;
  454. }
  455. msleep(1);
  456. }
  457. ipoib_dbg(priv, "All sends and receives done.\n");
  458. timeout:
  459. qp_attr.qp_state = IB_QPS_RESET;
  460. if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
  461. ipoib_warn(priv, "Failed to modify QP to RESET state\n");
  462. /* Wait for all AHs to be reaped */
  463. set_bit(IPOIB_STOP_REAPER, &priv->flags);
  464. cancel_delayed_work(&priv->ah_reap_task);
  465. flush_workqueue(ipoib_workqueue);
  466. begin = jiffies;
  467. while (!list_empty(&priv->dead_ahs)) {
  468. __ipoib_reap_ah(dev);
  469. if (time_after(jiffies, begin + HZ)) {
  470. ipoib_warn(priv, "timing out; will leak address handles\n");
  471. break;
  472. }
  473. msleep(1);
  474. }
  475. return 0;
  476. }
  477. int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
  478. {
  479. struct ipoib_dev_priv *priv = netdev_priv(dev);
  480. priv->ca = ca;
  481. priv->port = port;
  482. priv->qp = NULL;
  483. if (ipoib_transport_dev_init(dev, ca)) {
  484. printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name);
  485. return -ENODEV;
  486. }
  487. if (dev->flags & IFF_UP) {
  488. if (ipoib_ib_dev_open(dev)) {
  489. ipoib_transport_dev_cleanup(dev);
  490. return -ENODEV;
  491. }
  492. }
  493. return 0;
  494. }
  495. void ipoib_ib_dev_flush(void *_dev)
  496. {
  497. struct net_device *dev = (struct net_device *)_dev;
  498. struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv;
  499. if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags) ) {
  500. ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n");
  501. return;
  502. }
  503. if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
  504. ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n");
  505. return;
  506. }
  507. ipoib_dbg(priv, "flushing\n");
  508. ipoib_ib_dev_down(dev, 0);
  509. /*
  510. * The device could have been brought down between the start and when
  511. * we get here, don't bring it back up if it's not configured up
  512. */
  513. if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
  514. ipoib_ib_dev_up(dev);
  515. ipoib_mcast_restart_task(dev);
  516. }
  517. mutex_lock(&priv->vlan_mutex);
  518. /* Flush any child interfaces too */
  519. list_for_each_entry(cpriv, &priv->child_intfs, list)
  520. ipoib_ib_dev_flush(cpriv->dev);
  521. mutex_unlock(&priv->vlan_mutex);
  522. }
  523. void ipoib_ib_dev_cleanup(struct net_device *dev)
  524. {
  525. struct ipoib_dev_priv *priv = netdev_priv(dev);
  526. ipoib_dbg(priv, "cleaning up ib_dev\n");
  527. ipoib_mcast_stop_thread(dev, 1);
  528. ipoib_mcast_dev_flush(dev);
  529. ipoib_transport_dev_cleanup(dev);
  530. }
  531. /*
  532. * Delayed P_Key Assigment Interim Support
  533. *
  534. * The following is initial implementation of delayed P_Key assigment
  535. * mechanism. It is using the same approach implemented for the multicast
  536. * group join. The single goal of this implementation is to quickly address
  537. * Bug #2507. This implementation will probably be removed when the P_Key
  538. * change async notification is available.
  539. */
  540. void ipoib_pkey_poll(void *dev_ptr)
  541. {
  542. struct net_device *dev = dev_ptr;
  543. struct ipoib_dev_priv *priv = netdev_priv(dev);
  544. ipoib_pkey_dev_check_presence(dev);
  545. if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
  546. ipoib_open(dev);
  547. else {
  548. mutex_lock(&pkey_mutex);
  549. if (!test_bit(IPOIB_PKEY_STOP, &priv->flags))
  550. queue_delayed_work(ipoib_workqueue,
  551. &priv->pkey_task,
  552. HZ);
  553. mutex_unlock(&pkey_mutex);
  554. }
  555. }
  556. int ipoib_pkey_dev_delay_open(struct net_device *dev)
  557. {
  558. struct ipoib_dev_priv *priv = netdev_priv(dev);
  559. /* Look for the interface pkey value in the IB Port P_Key table and */
  560. /* set the interface pkey assigment flag */
  561. ipoib_pkey_dev_check_presence(dev);
  562. /* P_Key value not assigned yet - start polling */
  563. if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
  564. mutex_lock(&pkey_mutex);
  565. clear_bit(IPOIB_PKEY_STOP, &priv->flags);
  566. queue_delayed_work(ipoib_workqueue,
  567. &priv->pkey_task,
  568. HZ);
  569. mutex_unlock(&pkey_mutex);
  570. return 1;
  571. }
  572. return 0;
  573. }