2 * Copyright (c) 2006 Oracle. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/inetdevice.h>
38 #include <linux/if_arp.h>
39 #include <linux/delay.h>
40 #include <linux/slab.h>
45 unsigned int fmr_pool_size = RDS_FMR_POOL_SIZE;
46 unsigned int fmr_message_size = RDS_FMR_SIZE + 1; /* +1 allows for unaligned MRs */
47 unsigned int rds_ib_retry_count = RDS_IB_DEFAULT_RETRY_COUNT;
49 module_param(fmr_pool_size, int, 0444);
50 MODULE_PARM_DESC(fmr_pool_size, " Max number of fmr per HCA");
51 module_param(fmr_message_size, int, 0444);
52 MODULE_PARM_DESC(fmr_message_size, " Max size of a RDMA transfer");
53 module_param(rds_ib_retry_count, int, 0444);
54 MODULE_PARM_DESC(rds_ib_retry_count, " Number of hw retries before reporting an error");
56 struct list_head rds_ib_devices;
58 /* NOTE: if also grabbing ibdev lock, grab this first */
59 DEFINE_SPINLOCK(ib_nodev_conns_lock);
60 LIST_HEAD(ib_nodev_conns);
62 void rds_ib_add_one(struct ib_device *device)
64 struct rds_ib_device *rds_ibdev;
65 struct ib_device_attr *dev_attr;
67 /* Only handle IB (no iWARP) devices */
68 if (device->node_type != RDMA_NODE_IB_CA)
71 dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
75 if (ib_query_device(device, dev_attr)) {
76 rdsdebug("Query device failed for %s\n", device->name);
80 rds_ibdev = kmalloc(sizeof *rds_ibdev, GFP_KERNEL);
84 spin_lock_init(&rds_ibdev->spinlock);
86 rds_ibdev->max_wrs = dev_attr->max_qp_wr;
87 rds_ibdev->max_sge = min(dev_attr->max_sge, RDS_IB_MAX_SGE);
89 rds_ibdev->fmr_max_remaps = dev_attr->max_map_per_fmr?: 32;
90 rds_ibdev->max_fmrs = dev_attr->max_fmr ?
91 min_t(unsigned int, dev_attr->max_fmr, fmr_pool_size) :
94 rds_ibdev->max_initiator_depth = dev_attr->max_qp_init_rd_atom;
95 rds_ibdev->max_responder_resources = dev_attr->max_qp_rd_atom;
97 rds_ibdev->dev = device;
98 rds_ibdev->pd = ib_alloc_pd(device);
99 if (IS_ERR(rds_ibdev->pd))
102 rds_ibdev->mr = ib_get_dma_mr(rds_ibdev->pd,
103 IB_ACCESS_LOCAL_WRITE);
104 if (IS_ERR(rds_ibdev->mr))
107 rds_ibdev->mr_pool = rds_ib_create_mr_pool(rds_ibdev);
108 if (IS_ERR(rds_ibdev->mr_pool)) {
109 rds_ibdev->mr_pool = NULL;
113 INIT_LIST_HEAD(&rds_ibdev->ipaddr_list);
114 INIT_LIST_HEAD(&rds_ibdev->conn_list);
115 list_add_tail(&rds_ibdev->list, &rds_ib_devices);
117 ib_set_client_data(device, &rds_ib_client, rds_ibdev);
122 ib_dereg_mr(rds_ibdev->mr);
124 ib_dealloc_pd(rds_ibdev->pd);
131 void rds_ib_remove_one(struct ib_device *device)
133 struct rds_ib_device *rds_ibdev;
134 struct rds_ib_ipaddr *i_ipaddr, *i_next;
136 rds_ibdev = ib_get_client_data(device, &rds_ib_client);
140 list_for_each_entry_safe(i_ipaddr, i_next, &rds_ibdev->ipaddr_list, list) {
141 list_del(&i_ipaddr->list);
145 rds_ib_destroy_conns(rds_ibdev);
147 if (rds_ibdev->mr_pool)
148 rds_ib_destroy_mr_pool(rds_ibdev->mr_pool);
150 ib_dereg_mr(rds_ibdev->mr);
152 while (ib_dealloc_pd(rds_ibdev->pd)) {
153 rdsdebug("Failed to dealloc pd %p\n", rds_ibdev->pd);
157 list_del(&rds_ibdev->list);
161 struct ib_client rds_ib_client = {
163 .add = rds_ib_add_one,
164 .remove = rds_ib_remove_one
167 static int rds_ib_conn_info_visitor(struct rds_connection *conn,
170 struct rds_info_rdma_connection *iinfo = buffer;
171 struct rds_ib_connection *ic;
173 /* We will only ever look at IB transports */
174 if (conn->c_trans != &rds_ib_transport)
177 iinfo->src_addr = conn->c_laddr;
178 iinfo->dst_addr = conn->c_faddr;
180 memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid));
181 memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid));
182 if (rds_conn_state(conn) == RDS_CONN_UP) {
183 struct rds_ib_device *rds_ibdev;
184 struct rdma_dev_addr *dev_addr;
186 ic = conn->c_transport_data;
187 dev_addr = &ic->i_cm_id->route.addr.dev_addr;
189 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &iinfo->src_gid);
190 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &iinfo->dst_gid);
192 rds_ibdev = ib_get_client_data(ic->i_cm_id->device, &rds_ib_client);
193 iinfo->max_send_wr = ic->i_send_ring.w_nr;
194 iinfo->max_recv_wr = ic->i_recv_ring.w_nr;
195 iinfo->max_send_sge = rds_ibdev->max_sge;
196 rds_ib_get_mr_info(rds_ibdev, iinfo);
201 static void rds_ib_ic_info(struct socket *sock, unsigned int len,
202 struct rds_info_iterator *iter,
203 struct rds_info_lengths *lens)
205 rds_for_each_conn_info(sock, len, iter, lens,
206 rds_ib_conn_info_visitor,
207 sizeof(struct rds_info_rdma_connection));
212 * Early RDS/IB was built to only bind to an address if there is an IPoIB
213 * device with that address set.
215 * If it were me, I'd advocate for something more flexible. Sending and
216 * receiving should be device-agnostic. Transports would try and maintain
217 * connections between peers who have messages queued. Userspace would be
218 * allowed to influence which paths have priority. We could call userspace
219 * asserting this policy "routing".
221 static int rds_ib_laddr_check(__be32 addr)
224 struct rdma_cm_id *cm_id;
225 struct sockaddr_in sin;
227 /* Create a CMA ID and try to bind it. This catches both
228 * IB and iWARP capable NICs.
230 cm_id = rdma_create_id(NULL, NULL, RDMA_PS_TCP);
232 return PTR_ERR(cm_id);
234 memset(&sin, 0, sizeof(sin));
235 sin.sin_family = AF_INET;
236 sin.sin_addr.s_addr = addr;
238 /* rdma_bind_addr will only succeed for IB & iWARP devices */
239 ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin);
240 /* due to this, we will claim to support iWARP devices unless we
242 if (ret || cm_id->device->node_type != RDMA_NODE_IB_CA)
243 ret = -EADDRNOTAVAIL;
245 rdsdebug("addr %pI4 ret %d node type %d\n",
247 cm_id->device ? cm_id->device->node_type : -1);
249 rdma_destroy_id(cm_id);
254 void rds_ib_exit(void)
256 rds_info_deregister_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
257 rds_ib_destroy_nodev_conns();
258 ib_unregister_client(&rds_ib_client);
259 rds_ib_sysctl_exit();
261 rds_trans_unregister(&rds_ib_transport);
264 struct rds_transport rds_ib_transport = {
265 .laddr_check = rds_ib_laddr_check,
266 .xmit_complete = rds_ib_xmit_complete,
268 .xmit_rdma = rds_ib_xmit_rdma,
269 .xmit_atomic = rds_ib_xmit_atomic,
271 .conn_alloc = rds_ib_conn_alloc,
272 .conn_free = rds_ib_conn_free,
273 .conn_connect = rds_ib_conn_connect,
274 .conn_shutdown = rds_ib_conn_shutdown,
275 .inc_copy_to_user = rds_ib_inc_copy_to_user,
276 .inc_free = rds_ib_inc_free,
277 .cm_initiate_connect = rds_ib_cm_initiate_connect,
278 .cm_handle_connect = rds_ib_cm_handle_connect,
279 .cm_connect_complete = rds_ib_cm_connect_complete,
280 .stats_info_copy = rds_ib_stats_info_copy,
282 .get_mr = rds_ib_get_mr,
283 .sync_mr = rds_ib_sync_mr,
284 .free_mr = rds_ib_free_mr,
285 .flush_mrs = rds_ib_flush_mrs,
286 .t_owner = THIS_MODULE,
287 .t_name = "infiniband",
288 .t_type = RDS_TRANS_IB
291 int __init rds_ib_init(void)
295 INIT_LIST_HEAD(&rds_ib_devices);
297 ret = ib_register_client(&rds_ib_client);
301 ret = rds_ib_sysctl_init();
305 ret = rds_ib_recv_init();
309 ret = rds_trans_register(&rds_ib_transport);
313 rds_info_register_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
320 rds_ib_sysctl_exit();
322 ib_unregister_client(&rds_ib_client);
327 MODULE_LICENSE("GPL");