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>
44 unsigned int fmr_pool_size = RDS_FMR_POOL_SIZE;
45 unsigned int fmr_message_size = RDS_FMR_SIZE + 1; /* +1 allows for unaligned MRs */
47 module_param(fmr_pool_size, int, 0444);
48 MODULE_PARM_DESC(fmr_pool_size, " Max number of fmr per HCA");
49 module_param(fmr_message_size, int, 0444);
50 MODULE_PARM_DESC(fmr_message_size, " Max size of a RDMA transfer");
52 struct list_head rds_ib_devices;
54 DEFINE_SPINLOCK(ib_nodev_conns_lock);
55 LIST_HEAD(ib_nodev_conns);
57 void rds_ib_add_one(struct ib_device *device)
59 struct rds_ib_device *rds_ibdev;
60 struct ib_device_attr *dev_attr;
62 /* Only handle IB (no iWARP) devices */
63 if (device->node_type != RDMA_NODE_IB_CA)
66 dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
70 if (ib_query_device(device, dev_attr)) {
71 rdsdebug("Query device failed for %s\n", device->name);
75 rds_ibdev = kmalloc(sizeof *rds_ibdev, GFP_KERNEL);
79 spin_lock_init(&rds_ibdev->spinlock);
81 rds_ibdev->max_wrs = dev_attr->max_qp_wr;
82 rds_ibdev->max_sge = min(dev_attr->max_sge, RDS_IB_MAX_SGE);
84 rds_ibdev->fmr_page_shift = max(9, ffs(dev_attr->page_size_cap) - 1);
85 rds_ibdev->fmr_page_size = 1 << rds_ibdev->fmr_page_shift;
86 rds_ibdev->fmr_page_mask = ~((u64) rds_ibdev->fmr_page_size - 1);
87 rds_ibdev->fmr_max_remaps = dev_attr->max_map_per_fmr?: 32;
88 rds_ibdev->max_fmrs = dev_attr->max_fmr ?
89 min_t(unsigned int, dev_attr->max_fmr, fmr_pool_size) :
92 rds_ibdev->dev = device;
93 rds_ibdev->pd = ib_alloc_pd(device);
94 if (IS_ERR(rds_ibdev->pd))
97 rds_ibdev->mr = ib_get_dma_mr(rds_ibdev->pd,
98 IB_ACCESS_LOCAL_WRITE);
99 if (IS_ERR(rds_ibdev->mr))
102 rds_ibdev->mr_pool = rds_ib_create_mr_pool(rds_ibdev);
103 if (IS_ERR(rds_ibdev->mr_pool)) {
104 rds_ibdev->mr_pool = NULL;
108 INIT_LIST_HEAD(&rds_ibdev->ipaddr_list);
109 INIT_LIST_HEAD(&rds_ibdev->conn_list);
110 list_add_tail(&rds_ibdev->list, &rds_ib_devices);
112 ib_set_client_data(device, &rds_ib_client, rds_ibdev);
117 ib_dereg_mr(rds_ibdev->mr);
119 ib_dealloc_pd(rds_ibdev->pd);
126 void rds_ib_remove_one(struct ib_device *device)
128 struct rds_ib_device *rds_ibdev;
129 struct rds_ib_ipaddr *i_ipaddr, *i_next;
131 rds_ibdev = ib_get_client_data(device, &rds_ib_client);
135 list_for_each_entry_safe(i_ipaddr, i_next, &rds_ibdev->ipaddr_list, list) {
136 list_del(&i_ipaddr->list);
140 rds_ib_remove_conns(rds_ibdev);
142 if (rds_ibdev->mr_pool)
143 rds_ib_destroy_mr_pool(rds_ibdev->mr_pool);
145 ib_dereg_mr(rds_ibdev->mr);
147 while (ib_dealloc_pd(rds_ibdev->pd)) {
148 rdsdebug("Failed to dealloc pd %p\n", rds_ibdev->pd);
152 list_del(&rds_ibdev->list);
156 struct ib_client rds_ib_client = {
158 .add = rds_ib_add_one,
159 .remove = rds_ib_remove_one
162 static int rds_ib_conn_info_visitor(struct rds_connection *conn,
165 struct rds_info_rdma_connection *iinfo = buffer;
166 struct rds_ib_connection *ic;
168 /* We will only ever look at IB transports */
169 if (conn->c_trans != &rds_ib_transport)
172 iinfo->src_addr = conn->c_laddr;
173 iinfo->dst_addr = conn->c_faddr;
175 memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid));
176 memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid));
177 if (rds_conn_state(conn) == RDS_CONN_UP) {
178 struct rds_ib_device *rds_ibdev;
179 struct rdma_dev_addr *dev_addr;
181 ic = conn->c_transport_data;
182 dev_addr = &ic->i_cm_id->route.addr.dev_addr;
184 ib_addr_get_sgid(dev_addr, (union ib_gid *) &iinfo->src_gid);
185 ib_addr_get_dgid(dev_addr, (union ib_gid *) &iinfo->dst_gid);
187 rds_ibdev = ib_get_client_data(ic->i_cm_id->device, &rds_ib_client);
188 iinfo->max_send_wr = ic->i_send_ring.w_nr;
189 iinfo->max_recv_wr = ic->i_recv_ring.w_nr;
190 iinfo->max_send_sge = rds_ibdev->max_sge;
191 rds_ib_get_mr_info(rds_ibdev, iinfo);
196 static void rds_ib_ic_info(struct socket *sock, unsigned int len,
197 struct rds_info_iterator *iter,
198 struct rds_info_lengths *lens)
200 rds_for_each_conn_info(sock, len, iter, lens,
201 rds_ib_conn_info_visitor,
202 sizeof(struct rds_info_rdma_connection));
207 * Early RDS/IB was built to only bind to an address if there is an IPoIB
208 * device with that address set.
210 * If it were me, I'd advocate for something more flexible. Sending and
211 * receiving should be device-agnostic. Transports would try and maintain
212 * connections between peers who have messages queued. Userspace would be
213 * allowed to influence which paths have priority. We could call userspace
214 * asserting this policy "routing".
216 static int rds_ib_laddr_check(__be32 addr)
219 struct rdma_cm_id *cm_id;
220 struct sockaddr_in sin;
222 /* Create a CMA ID and try to bind it. This catches both
223 * IB and iWARP capable NICs.
225 cm_id = rdma_create_id(NULL, NULL, RDMA_PS_TCP);
227 return -EADDRNOTAVAIL;
229 memset(&sin, 0, sizeof(sin));
230 sin.sin_family = AF_INET;
231 sin.sin_addr.s_addr = addr;
233 /* rdma_bind_addr will only succeed for IB & iWARP devices */
234 ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin);
235 /* due to this, we will claim to support iWARP devices unless we
237 if (ret || cm_id->device->node_type != RDMA_NODE_IB_CA)
238 ret = -EADDRNOTAVAIL;
240 rdsdebug("addr %pI4 ret %d node type %d\n",
242 cm_id->device ? cm_id->device->node_type : -1);
244 rdma_destroy_id(cm_id);
249 void rds_ib_exit(void)
251 rds_info_deregister_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
252 rds_ib_remove_nodev_conns();
253 ib_unregister_client(&rds_ib_client);
254 rds_ib_sysctl_exit();
256 rds_trans_unregister(&rds_ib_transport);
259 struct rds_transport rds_ib_transport = {
260 .laddr_check = rds_ib_laddr_check,
261 .xmit_complete = rds_ib_xmit_complete,
263 .xmit_cong_map = NULL,
264 .xmit_rdma = rds_ib_xmit_rdma,
266 .conn_alloc = rds_ib_conn_alloc,
267 .conn_free = rds_ib_conn_free,
268 .conn_connect = rds_ib_conn_connect,
269 .conn_shutdown = rds_ib_conn_shutdown,
270 .inc_copy_to_user = rds_ib_inc_copy_to_user,
271 .inc_purge = rds_ib_inc_purge,
272 .inc_free = rds_ib_inc_free,
273 .cm_initiate_connect = rds_ib_cm_initiate_connect,
274 .cm_handle_connect = rds_ib_cm_handle_connect,
275 .cm_connect_complete = rds_ib_cm_connect_complete,
276 .stats_info_copy = rds_ib_stats_info_copy,
278 .get_mr = rds_ib_get_mr,
279 .sync_mr = rds_ib_sync_mr,
280 .free_mr = rds_ib_free_mr,
281 .flush_mrs = rds_ib_flush_mrs,
282 .t_owner = THIS_MODULE,
283 .t_name = "infiniband",
286 int __init rds_ib_init(void)
290 INIT_LIST_HEAD(&rds_ib_devices);
292 ret = ib_register_client(&rds_ib_client);
296 ret = rds_ib_sysctl_init();
300 ret = rds_ib_recv_init();
304 ret = rds_trans_register(&rds_ib_transport);
308 rds_info_register_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
315 rds_ib_sysctl_exit();
317 ib_unregister_client(&rds_ib_client);
322 MODULE_LICENSE("GPL");