ipoib_verbs.c 8.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314
  1. /*
  2. * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
  3. * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
  4. *
  5. * This software is available to you under a choice of one of two
  6. * licenses. You may choose to be licensed under the terms of the GNU
  7. * General Public License (GPL) Version 2, available from the file
  8. * COPYING in the main directory of this source tree, or the
  9. * OpenIB.org BSD license below:
  10. *
  11. * Redistribution and use in source and binary forms, with or
  12. * without modification, are permitted provided that the following
  13. * conditions are met:
  14. *
  15. * - Redistributions of source code must retain the above
  16. * copyright notice, this list of conditions and the following
  17. * disclaimer.
  18. *
  19. * - Redistributions in binary form must reproduce the above
  20. * copyright notice, this list of conditions and the following
  21. * disclaimer in the documentation and/or other materials
  22. * provided with the distribution.
  23. *
  24. * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25. * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26. * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27. * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28. * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29. * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30. * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31. * SOFTWARE.
  32. */
  33. #include <linux/slab.h>
  34. #include "ipoib.h"
  35. int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid, int set_qkey)
  36. {
  37. struct ipoib_dev_priv *priv = netdev_priv(dev);
  38. struct ib_qp_attr *qp_attr = NULL;
  39. int ret;
  40. u16 pkey_index;
  41. if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) {
  42. clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
  43. ret = -ENXIO;
  44. goto out;
  45. }
  46. set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
  47. if (set_qkey) {
  48. ret = -ENOMEM;
  49. qp_attr = kmalloc(sizeof *qp_attr, GFP_KERNEL);
  50. if (!qp_attr)
  51. goto out;
  52. /* set correct QKey for QP */
  53. qp_attr->qkey = priv->qkey;
  54. ret = ib_modify_qp(priv->qp, qp_attr, IB_QP_QKEY);
  55. if (ret) {
  56. ipoib_warn(priv, "failed to modify QP, ret = %d\n", ret);
  57. goto out;
  58. }
  59. }
  60. /* attach QP to multicast group */
  61. ret = ib_attach_mcast(priv->qp, mgid, mlid);
  62. if (ret)
  63. ipoib_warn(priv, "failed to attach to multicast group, ret = %d\n", ret);
  64. out:
  65. kfree(qp_attr);
  66. return ret;
  67. }
  68. int ipoib_init_qp(struct net_device *dev)
  69. {
  70. struct ipoib_dev_priv *priv = netdev_priv(dev);
  71. int ret;
  72. struct ib_qp_attr qp_attr;
  73. int attr_mask;
  74. if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
  75. return -1;
  76. qp_attr.qp_state = IB_QPS_INIT;
  77. qp_attr.qkey = 0;
  78. qp_attr.port_num = priv->port;
  79. qp_attr.pkey_index = priv->pkey_index;
  80. attr_mask =
  81. IB_QP_QKEY |
  82. IB_QP_PORT |
  83. IB_QP_PKEY_INDEX |
  84. IB_QP_STATE;
  85. ret = ib_modify_qp(priv->qp, &qp_attr, attr_mask);
  86. if (ret) {
  87. ipoib_warn(priv, "failed to modify QP to init, ret = %d\n", ret);
  88. goto out_fail;
  89. }
  90. qp_attr.qp_state = IB_QPS_RTR;
  91. /* Can't set this in a INIT->RTR transition */
  92. attr_mask &= ~IB_QP_PORT;
  93. ret = ib_modify_qp(priv->qp, &qp_attr, attr_mask);
  94. if (ret) {
  95. ipoib_warn(priv, "failed to modify QP to RTR, ret = %d\n", ret);
  96. goto out_fail;
  97. }
  98. qp_attr.qp_state = IB_QPS_RTS;
  99. qp_attr.sq_psn = 0;
  100. attr_mask |= IB_QP_SQ_PSN;
  101. attr_mask &= ~IB_QP_PKEY_INDEX;
  102. ret = ib_modify_qp(priv->qp, &qp_attr, attr_mask);
  103. if (ret) {
  104. ipoib_warn(priv, "failed to modify QP to RTS, ret = %d\n", ret);
  105. goto out_fail;
  106. }
  107. return 0;
  108. out_fail:
  109. qp_attr.qp_state = IB_QPS_RESET;
  110. if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
  111. ipoib_warn(priv, "Failed to modify QP to RESET state\n");
  112. return ret;
  113. }
  114. int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
  115. {
  116. struct ipoib_dev_priv *priv = netdev_priv(dev);
  117. struct ib_qp_init_attr init_attr = {
  118. .cap = {
  119. .max_send_wr = ipoib_sendq_size,
  120. .max_recv_wr = ipoib_recvq_size,
  121. .max_send_sge = min_t(u32, priv->ca->attrs.max_sge,
  122. MAX_SKB_FRAGS + 1),
  123. .max_recv_sge = IPOIB_UD_RX_SG
  124. },
  125. .sq_sig_type = IB_SIGNAL_ALL_WR,
  126. .qp_type = IB_QPT_UD
  127. };
  128. struct ib_cq_init_attr cq_attr = {};
  129. int ret, size;
  130. int i;
  131. priv->pd = ib_alloc_pd(priv->ca, 0);
  132. if (IS_ERR(priv->pd)) {
  133. printk(KERN_WARNING "%s: failed to allocate PD\n", ca->name);
  134. return -ENODEV;
  135. }
  136. /*
  137. * the various IPoIB tasks assume they will never race against
  138. * themselves, so always use a single thread workqueue
  139. */
  140. priv->wq = alloc_ordered_workqueue("ipoib_wq", WQ_MEM_RECLAIM);
  141. if (!priv->wq) {
  142. printk(KERN_WARNING "ipoib: failed to allocate device WQ\n");
  143. goto out_free_pd;
  144. }
  145. size = ipoib_recvq_size + 1;
  146. ret = ipoib_cm_dev_init(dev);
  147. if (!ret) {
  148. size += ipoib_sendq_size;
  149. if (ipoib_cm_has_srq(dev))
  150. size += ipoib_recvq_size + 1; /* 1 extra for rx_drain_qp */
  151. else
  152. size += ipoib_recvq_size * ipoib_max_conn_qp;
  153. } else
  154. if (ret != -ENOSYS)
  155. goto out_free_wq;
  156. cq_attr.cqe = size;
  157. priv->recv_cq = ib_create_cq(priv->ca, ipoib_ib_completion, NULL,
  158. dev, &cq_attr);
  159. if (IS_ERR(priv->recv_cq)) {
  160. printk(KERN_WARNING "%s: failed to create receive CQ\n", ca->name);
  161. goto out_cm_dev_cleanup;
  162. }
  163. cq_attr.cqe = ipoib_sendq_size;
  164. priv->send_cq = ib_create_cq(priv->ca, ipoib_send_comp_handler, NULL,
  165. dev, &cq_attr);
  166. if (IS_ERR(priv->send_cq)) {
  167. printk(KERN_WARNING "%s: failed to create send CQ\n", ca->name);
  168. goto out_free_recv_cq;
  169. }
  170. if (ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP))
  171. goto out_free_send_cq;
  172. init_attr.send_cq = priv->send_cq;
  173. init_attr.recv_cq = priv->recv_cq;
  174. if (priv->hca_caps & IB_DEVICE_UD_TSO)
  175. init_attr.create_flags |= IB_QP_CREATE_IPOIB_UD_LSO;
  176. if (priv->hca_caps & IB_DEVICE_BLOCK_MULTICAST_LOOPBACK)
  177. init_attr.create_flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
  178. if (priv->hca_caps & IB_DEVICE_MANAGED_FLOW_STEERING)
  179. init_attr.create_flags |= IB_QP_CREATE_NETIF_QP;
  180. priv->qp = ib_create_qp(priv->pd, &init_attr);
  181. if (IS_ERR(priv->qp)) {
  182. printk(KERN_WARNING "%s: failed to create QP\n", ca->name);
  183. goto out_free_send_cq;
  184. }
  185. priv->dev->dev_addr[1] = (priv->qp->qp_num >> 16) & 0xff;
  186. priv->dev->dev_addr[2] = (priv->qp->qp_num >> 8) & 0xff;
  187. priv->dev->dev_addr[3] = (priv->qp->qp_num ) & 0xff;
  188. for (i = 0; i < MAX_SKB_FRAGS + 1; ++i)
  189. priv->tx_sge[i].lkey = priv->pd->local_dma_lkey;
  190. priv->tx_wr.wr.opcode = IB_WR_SEND;
  191. priv->tx_wr.wr.sg_list = priv->tx_sge;
  192. priv->tx_wr.wr.send_flags = IB_SEND_SIGNALED;
  193. priv->rx_sge[0].lkey = priv->pd->local_dma_lkey;
  194. priv->rx_sge[0].length = IPOIB_UD_BUF_SIZE(priv->max_ib_mtu);
  195. priv->rx_wr.num_sge = 1;
  196. priv->rx_wr.next = NULL;
  197. priv->rx_wr.sg_list = priv->rx_sge;
  198. if (init_attr.cap.max_send_sge > 1)
  199. dev->features |= NETIF_F_SG;
  200. priv->max_send_sge = init_attr.cap.max_send_sge;
  201. return 0;
  202. out_free_send_cq:
  203. ib_destroy_cq(priv->send_cq);
  204. out_free_recv_cq:
  205. ib_destroy_cq(priv->recv_cq);
  206. out_cm_dev_cleanup:
  207. ipoib_cm_dev_cleanup(dev);
  208. out_free_wq:
  209. destroy_workqueue(priv->wq);
  210. priv->wq = NULL;
  211. out_free_pd:
  212. ib_dealloc_pd(priv->pd);
  213. return -ENODEV;
  214. }
  215. void ipoib_transport_dev_cleanup(struct net_device *dev)
  216. {
  217. struct ipoib_dev_priv *priv = netdev_priv(dev);
  218. if (priv->qp) {
  219. if (ib_destroy_qp(priv->qp))
  220. ipoib_warn(priv, "ib_qp_destroy failed\n");
  221. priv->qp = NULL;
  222. clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
  223. }
  224. if (ib_destroy_cq(priv->send_cq))
  225. ipoib_warn(priv, "ib_cq_destroy (send) failed\n");
  226. if (ib_destroy_cq(priv->recv_cq))
  227. ipoib_warn(priv, "ib_cq_destroy (recv) failed\n");
  228. ipoib_cm_dev_cleanup(dev);
  229. if (priv->wq) {
  230. flush_workqueue(priv->wq);
  231. destroy_workqueue(priv->wq);
  232. priv->wq = NULL;
  233. }
  234. ib_dealloc_pd(priv->pd);
  235. }
  236. void ipoib_event(struct ib_event_handler *handler,
  237. struct ib_event *record)
  238. {
  239. struct ipoib_dev_priv *priv =
  240. container_of(handler, struct ipoib_dev_priv, event_handler);
  241. if (record->element.port_num != priv->port)
  242. return;
  243. ipoib_dbg(priv, "Event %d on device %s port %d\n", record->event,
  244. record->device->name, record->element.port_num);
  245. if (record->event == IB_EVENT_SM_CHANGE ||
  246. record->event == IB_EVENT_CLIENT_REREGISTER) {
  247. queue_work(ipoib_workqueue, &priv->flush_light);
  248. } else if (record->event == IB_EVENT_PORT_ERR ||
  249. record->event == IB_EVENT_PORT_ACTIVE ||
  250. record->event == IB_EVENT_LID_CHANGE) {
  251. queue_work(ipoib_workqueue, &priv->flush_normal);
  252. } else if (record->event == IB_EVENT_PKEY_CHANGE) {
  253. queue_work(ipoib_workqueue, &priv->flush_heavy);
  254. } else if (record->event == IB_EVENT_GID_CHANGE &&
  255. !test_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags)) {
  256. queue_work(ipoib_workqueue, &priv->flush_light);
  257. }
  258. }