2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
34 #include <linux/slab.h>
38 int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid, int set_qkey)
40 struct ipoib_dev_priv *priv = netdev_priv(dev);
41 struct ib_qp_attr *qp_attr = NULL;
45 if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &pkey_index)) {
46 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
50 set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
54 qp_attr = kmalloc(sizeof *qp_attr, GFP_KERNEL);
58 /* set correct QKey for QP */
59 qp_attr->qkey = priv->qkey;
60 ret = ib_modify_qp(priv->qp, qp_attr, IB_QP_QKEY);
62 ipoib_warn(priv, "failed to modify QP, ret = %d\n", ret);
67 /* attach QP to multicast group */
68 ret = ib_attach_mcast(priv->qp, mgid, mlid);
70 ipoib_warn(priv, "failed to attach to multicast group, ret = %d\n", ret);
77 int ipoib_init_qp(struct net_device *dev)
79 struct ipoib_dev_priv *priv = netdev_priv(dev);
81 struct ib_qp_attr qp_attr;
84 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
87 qp_attr.qp_state = IB_QPS_INIT;
89 qp_attr.port_num = priv->port;
90 qp_attr.pkey_index = priv->pkey_index;
96 ret = ib_modify_qp(priv->qp, &qp_attr, attr_mask);
98 ipoib_warn(priv, "failed to modify QP to init, ret = %d\n", ret);
102 qp_attr.qp_state = IB_QPS_RTR;
103 /* Can't set this in a INIT->RTR transition */
104 attr_mask &= ~IB_QP_PORT;
105 ret = ib_modify_qp(priv->qp, &qp_attr, attr_mask);
107 ipoib_warn(priv, "failed to modify QP to RTR, ret = %d\n", ret);
111 qp_attr.qp_state = IB_QPS_RTS;
113 attr_mask |= IB_QP_SQ_PSN;
114 attr_mask &= ~IB_QP_PKEY_INDEX;
115 ret = ib_modify_qp(priv->qp, &qp_attr, attr_mask);
117 ipoib_warn(priv, "failed to modify QP to RTS, ret = %d\n", ret);
124 qp_attr.qp_state = IB_QPS_RESET;
125 if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
126 ipoib_warn(priv, "Failed to modify QP to RESET state\n");
131 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
133 struct ipoib_dev_priv *priv = netdev_priv(dev);
134 struct ib_qp_init_attr init_attr = {
136 .max_send_wr = ipoib_sendq_size,
137 .max_recv_wr = ipoib_recvq_size,
139 .max_recv_sge = IPOIB_UD_RX_SG
141 .sq_sig_type = IB_SIGNAL_ALL_WR,
144 struct ib_cq_init_attr cq_attr = {};
149 priv->pd = ib_alloc_pd(priv->ca);
150 if (IS_ERR(priv->pd)) {
151 printk(KERN_WARNING "%s: failed to allocate PD\n", ca->name);
155 priv->mr = ib_get_dma_mr(priv->pd, IB_ACCESS_LOCAL_WRITE);
156 if (IS_ERR(priv->mr)) {
157 printk(KERN_WARNING "%s: ib_get_dma_mr failed\n", ca->name);
162 * the various IPoIB tasks assume they will never race against
163 * themselves, so always use a single thread workqueue
165 priv->wq = create_singlethread_workqueue("ipoib_wq");
167 printk(KERN_WARNING "ipoib: failed to allocate device WQ\n");
171 size = ipoib_recvq_size + 1;
172 ret = ipoib_cm_dev_init(dev);
174 size += ipoib_sendq_size;
175 if (ipoib_cm_has_srq(dev))
176 size += ipoib_recvq_size + 1; /* 1 extra for rx_drain_qp */
178 size += ipoib_recvq_size * ipoib_max_conn_qp;
183 priv->recv_cq = ib_create_cq(priv->ca, ipoib_ib_completion, NULL,
185 if (IS_ERR(priv->recv_cq)) {
186 printk(KERN_WARNING "%s: failed to create receive CQ\n", ca->name);
187 goto out_cm_dev_cleanup;
190 cq_attr.cqe = ipoib_sendq_size;
191 priv->send_cq = ib_create_cq(priv->ca, ipoib_send_comp_handler, NULL,
193 if (IS_ERR(priv->send_cq)) {
194 printk(KERN_WARNING "%s: failed to create send CQ\n", ca->name);
195 goto out_free_recv_cq;
198 if (ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP))
199 goto out_free_send_cq;
201 init_attr.send_cq = priv->send_cq;
202 init_attr.recv_cq = priv->recv_cq;
204 if (priv->hca_caps & IB_DEVICE_UD_TSO)
205 init_attr.create_flags |= IB_QP_CREATE_IPOIB_UD_LSO;
207 if (priv->hca_caps & IB_DEVICE_BLOCK_MULTICAST_LOOPBACK)
208 init_attr.create_flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
210 if (priv->hca_caps & IB_DEVICE_MANAGED_FLOW_STEERING)
211 init_attr.create_flags |= IB_QP_CREATE_NETIF_QP;
213 if (dev->features & NETIF_F_SG)
214 init_attr.cap.max_send_sge = MAX_SKB_FRAGS + 1;
216 priv->qp = ib_create_qp(priv->pd, &init_attr);
217 if (IS_ERR(priv->qp)) {
218 printk(KERN_WARNING "%s: failed to create QP\n", ca->name);
219 goto out_free_send_cq;
222 priv->dev->dev_addr[1] = (priv->qp->qp_num >> 16) & 0xff;
223 priv->dev->dev_addr[2] = (priv->qp->qp_num >> 8) & 0xff;
224 priv->dev->dev_addr[3] = (priv->qp->qp_num ) & 0xff;
226 for (i = 0; i < MAX_SKB_FRAGS + 1; ++i)
227 priv->tx_sge[i].lkey = priv->mr->lkey;
229 priv->tx_wr.opcode = IB_WR_SEND;
230 priv->tx_wr.sg_list = priv->tx_sge;
231 priv->tx_wr.send_flags = IB_SEND_SIGNALED;
233 priv->rx_sge[0].lkey = priv->mr->lkey;
235 priv->rx_sge[0].length = IPOIB_UD_BUF_SIZE(priv->max_ib_mtu);
236 priv->rx_wr.num_sge = 1;
238 priv->rx_wr.next = NULL;
239 priv->rx_wr.sg_list = priv->rx_sge;
244 ib_destroy_cq(priv->send_cq);
247 ib_destroy_cq(priv->recv_cq);
250 ipoib_cm_dev_cleanup(dev);
253 destroy_workqueue(priv->wq);
257 ib_dereg_mr(priv->mr);
260 ib_dealloc_pd(priv->pd);
265 void ipoib_transport_dev_cleanup(struct net_device *dev)
267 struct ipoib_dev_priv *priv = netdev_priv(dev);
270 if (ib_destroy_qp(priv->qp))
271 ipoib_warn(priv, "ib_qp_destroy failed\n");
274 clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
277 if (ib_destroy_cq(priv->send_cq))
278 ipoib_warn(priv, "ib_cq_destroy (send) failed\n");
280 if (ib_destroy_cq(priv->recv_cq))
281 ipoib_warn(priv, "ib_cq_destroy (recv) failed\n");
283 ipoib_cm_dev_cleanup(dev);
286 flush_workqueue(priv->wq);
287 destroy_workqueue(priv->wq);
291 if (ib_dereg_mr(priv->mr))
292 ipoib_warn(priv, "ib_dereg_mr failed\n");
294 if (ib_dealloc_pd(priv->pd))
295 ipoib_warn(priv, "ib_dealloc_pd failed\n");
299 void ipoib_event(struct ib_event_handler *handler,
300 struct ib_event *record)
302 struct ipoib_dev_priv *priv =
303 container_of(handler, struct ipoib_dev_priv, event_handler);
305 if (record->element.port_num != priv->port)
308 ipoib_dbg(priv, "Event %d on device %s port %d\n", record->event,
309 record->device->name, record->element.port_num);
311 if (record->event == IB_EVENT_SM_CHANGE ||
312 record->event == IB_EVENT_CLIENT_REREGISTER) {
313 queue_work(ipoib_workqueue, &priv->flush_light);
314 } else if (record->event == IB_EVENT_PORT_ERR ||
315 record->event == IB_EVENT_PORT_ACTIVE ||
316 record->event == IB_EVENT_LID_CHANGE) {
317 queue_work(ipoib_workqueue, &priv->flush_normal);
318 } else if (record->event == IB_EVENT_PKEY_CHANGE) {
319 queue_work(ipoib_workqueue, &priv->flush_heavy);